├── .gitignore ├── LICENSE ├── README.md ├── analyze.py ├── compare.py ├── metrics.py ├── ml ├── grid_search │ ├── AdaBoostClassifier.py │ ├── BernoulliNB.py │ ├── DecisionTreeClassifier.py │ ├── ExtraTreesClassifier.py │ ├── GaussianNB.py │ ├── GradientBoostingClassifier.py │ ├── KNeighborsClassifier.py │ ├── LinearSVC.py │ ├── LogisticRegression.py │ ├── MultinomialNB.py │ ├── PassiveAggressiveClassifier.py │ ├── RandomForestClassifier.py │ ├── SGDClassifier.py │ ├── SVC.py │ ├── XGBClassifier.py │ ├── evaluate_model.py │ └── tpot_metrics.py ├── random_search │ ├── AdaBoostClassifier.py │ ├── BernoulliNB.py │ ├── DecisionTreeClassifier.py │ ├── ExtraTreesClassifier.py │ ├── GaussianNB.py │ ├── GradientBoostingClassifier.py │ ├── KNeighborsClassifier.py │ ├── LinearSVC.py │ ├── LogisticRegression.py │ ├── MultinomialNB.py │ ├── PassiveAggressiveClassifier.py │ ├── RandomForestClassifier.py │ ├── SGDClassifier.py │ ├── SVC.py │ ├── XGBClassifier.py │ ├── evaluate_model.py │ └── tpot_metrics.py └── random_search_preprocessing │ ├── AdaBoostClassifier.py │ ├── BernoulliNB.py │ ├── DecisionTreeClassifier.py │ ├── ExtraTreesClassifier.py │ ├── GaussianNB.py │ ├── GradientBoostingClassifier.py │ ├── KNeighborsClassifier.py │ ├── LinearSVC.py │ ├── LogisticRegression.py │ ├── MLPClassifier.py │ ├── MultinomialNB.py │ ├── PassiveAggressiveClassifier.py │ ├── RandomForestClassifier.py │ ├── SGDClassifier.py │ ├── SVC.py │ ├── XGBClassifier.py │ ├── evaluate_model.py │ └── tpot_metrics.py ├── preprocessors.py ├── read_file.py ├── submit_jobs.py └── utils.py /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | env/ 12 | build/ 13 | develop-eggs/ 14 | dist/ 15 | downloads/ 16 | eggs/ 17 | .eggs/ 18 | lib/ 19 | lib64/ 20 | parts/ 21 | sdist/ 22 | var/ 23 | wheels/ 24 | *.egg-info/ 25 | .installed.cfg 26 | *.egg 27 | 28 | # PyInstaller 29 | # Usually these files are written by a python script from a template 30 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 31 | *.manifest 32 | *.spec 33 | 34 | # Installer logs 35 | pip-log.txt 36 | pip-delete-this-directory.txt 37 | 38 | # Unit test / coverage reports 39 | htmlcov/ 40 | .tox/ 41 | .coverage 42 | .coverage.* 43 | .cache 44 | nosetests.xml 45 | coverage.xml 46 | *.cover 47 | .hypothesis/ 48 | 49 | # Translations 50 | *.mo 51 | *.pot 52 | 53 | # Django stuff: 54 | *.log 55 | local_settings.py 56 | 57 | # Flask stuff: 58 | instance/ 59 | .webassets-cache 60 | 61 | # Scrapy stuff: 62 | .scrapy 63 | 64 | # Sphinx documentation 65 | docs/_build/ 66 | 67 | # PyBuilder 68 | target/ 69 | 70 | # Jupyter Notebook 71 | .ipynb_checkpoints 72 | 73 | # pyenv 74 | .python-version 75 | 76 | # celery beat schedule file 77 | celerybeat-schedule 78 | 79 | # SageMath parsed files 80 | *.sage.py 81 | 82 | # dotenv 83 | .env 84 | 85 | # virtualenv 86 | .venv 87 | venv/ 88 | ENV/ 89 | 90 | # Spyder project settings 91 | .spyderproject 92 | .spyproject 93 | 94 | # Rope project settings 95 | .ropeproject 96 | 97 | # mkdocs documentation 98 | /site 99 | 100 | # mypy 101 | .mypy_cache/ 102 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # ML Analyst 2 | 3 | This project is designed to help rapidly apply a standard machine learning analysis to a new data set. 4 | 5 | 6 | # Usage 7 | 8 | ## running the analysis 9 | 10 | > I want to run logistic regression, random forests, and neural nets on my dataset. i want to scale the features. 11 | 12 | ```python 13 | python analyze.py path/to/dataset -ml LogisticRegression,RandomForestClassifier,MLPClassifier -prep RobustScaler 14 | ``` 15 | > I want to tune the parameters of each method using 100 combinations, and run 10 shuffles of the data. 16 | 17 | ```python 18 | python analyze.py path/to/dataset -ml LogisticRegression,RandomForestClassifier,MLPClassifier -prep RobustScaler -n_combos 100 -n_trials 10 19 | ``` 20 | 21 | > what other options are there? 22 | ``` 23 | python analyze.py -h 24 | ``` 25 | 26 | ## generating comparisons 27 | 28 | ```python 29 | python compare.py path/to/results 30 | ``` 31 | -------------------------------------------------------------------------------- /analyze.py: -------------------------------------------------------------------------------- 1 | import pandas as pd 2 | import numpy as np 3 | import argparse 4 | import os, errno, sys 5 | from sklearn.externals.joblib import Parallel, delayed 6 | 7 | 8 | if __name__ == '__main__': 9 | # parse command line arguments 10 | parser = argparse.ArgumentParser(description="An analyst for quick ML applications.", 11 | add_help=False) 12 | parser.add_argument('INPUT_FILE', type=str, 13 | help='Data file to analyze; ensure that the ' 14 | 'target/label column is labeled as "class".') 15 | parser.add_argument('-h', '--help', action='help', 16 | help='Show this help message and exit.') 17 | parser.add_argument('-ml', action='store', dest='LEARNERS',default=None,type=str, 18 | help='Comma-separated list of ML methods to use (should correspond to a py file name ' 19 | 'in ml/)') 20 | parser.add_argument('-prep', action='store', dest='PREP', default=None, type=str, 21 | help = 'Comma-separated list of preprocessors to apply to data') 22 | parser.add_argument('--lsf', action='store_true', dest='LSF', default=False, 23 | help='Run on an LSF HPC (using bsub commands)') 24 | parser.add_argument('-metric',action='store', dest='METRIC', default='f1_macro', type=str, 25 | help='Metric to compare algorithms') 26 | parser.add_argument('-k',action='store', dest='K', default=5, type=int, 27 | help='Number of folds for cross validation') 28 | parser.add_argument('-search',action='store',dest='SEARCH',default='random',choices=['grid','random'], 29 | help='Hyperparameter search strategy') 30 | parser.add_argument('--r',action='store_true',dest='REGRESSION',default=False, 31 | help='Run regression instead of classification.') 32 | parser.add_argument('-n_jobs',action='store',dest='N_JOBS',default=4,type=int, 33 | help='Number of parallel jobs') 34 | parser.add_argument('-n_trials',action='store',dest='N_TRIALS',default=1,type=int, 35 | help='Number of parallel jobs') 36 | parser.add_argument('-n_combos',action='store',dest='N_COMBOS',default=4,type=int, 37 | help='Number of hyperparameters to try') 38 | parser.add_argument('-rs',action='store',dest='RANDOM_STATE',default=None,type=int, 39 | help='random state') 40 | parser.add_argument('-label',action='store',dest='LABEL',default='class',type=str, 41 | help='Name of class label column') 42 | parser.add_argument('-m',action='store',dest='M',default=4096,type=int, 43 | help='LSF memory request and limit (MB)') 44 | parser.add_argument('-results',action='store',dest='RDIR',default='results',type=str, 45 | help='results directory') 46 | parser.add_argument('-q',action='store',dest='QUEUE',default='moore_normal',type=str, 47 | help='results directory') 48 | 49 | 50 | args = parser.parse_args() 51 | 52 | if args.RANDOM_STATE: 53 | random_state = args.RANDOM_STATE 54 | else: 55 | random_state = np.random.randint(2**15 - 1) 56 | 57 | learners = [ml for ml in args.LEARNERS.split(',')] # learners 58 | 59 | if args.SEARCH == 'random': 60 | if args.PREP: 61 | model_dir = 'ml/random_search_preprocessing/' 62 | else: 63 | model_dir = 'ml/random_search/' 64 | else: 65 | model_dir= 'ml/grid_search/' 66 | 67 | dataset = args.INPUT_FILE.split('/')[-1].split('.csv')[0] 68 | RANDOM_STATE = args.RANDOM_STATE 69 | results_path = '/'.join([args.RDIR, dataset]) + '/' 70 | # make the results_path directory if it doesn't exit 71 | try: 72 | os.makedirs(results_path) 73 | except OSError as e: 74 | if e.errno != errno.EEXIST: 75 | raise 76 | # initialize output files 77 | for ml in learners: 78 | #write headers 79 | if args.PREP: 80 | save_file = results_path + '-'.join(args.PREP.split(',')) + '_' + ml + '.csv' 81 | else: 82 | save_file = results_path + '/' + dataset + '_' + ml + '.csv' 83 | 84 | feat_file = save_file.split('.')[0]+'.imp_score' 85 | roc_file = save_file.split('.')[0]+'.roc' 86 | 87 | with open(save_file.split('.')[0] + '.imp_score','w') as out: 88 | out.write('preprocessor\tprep-parameters\talgorithm\talg-parameters\tseed\tfeature\tscore\n') 89 | 90 | with open(save_file.split('.')[0] + '.roc','w') as out: 91 | out.write('preprocessor\tprep-parameters\talgorithm\talg-parameters\tseed\tfpr\ttpr\tauc\n') 92 | 93 | with open(save_file,'w') as out: 94 | if args.PREP: 95 | out.write('dataset\tpreprocessor\tprep-parameters\talgorithm\talg-parameters\tseed\taccuracy\tf1_macro\tbal_accuracy\troc_auc\n') 96 | else: 97 | out.write('dataset\talgorithm\tparameters\taccuracy\tf1_macro\tseed\tbal_accuracy\troc_auc\n') 98 | 99 | # write run commands 100 | all_commands = [] 101 | job_info=[] 102 | for t in range(args.N_TRIALS): 103 | random_state = np.random.randint(2**15-1) 104 | for ml in learners: 105 | if args.PREP: 106 | save_file = results_path + '-'.join(args.PREP.split(',')) + '_' + ml + '.csv' 107 | else: 108 | save_file = results_path + '/' + dataset + '_' + ml + '.csv' 109 | 110 | if args.PREP: 111 | all_commands.append('python {PATH}/{ML}.py {DATASET} {SAVEFILE} {N_COMBOS} {RS} {PREP} {LABEL}'.format(PATH=model_dir,ML=ml,DATASET=args.INPUT_FILE,SAVEFILE=save_file,N_COMBOS=args.N_COMBOS,RS=random_state,PREP=args.PREP,LABEL=args.LABEL)) 112 | elif args.SEARCH == 'random': 113 | all_commands.append('python {PATH}/{ML}.py {DATASET} {SAVEFILE} {N_COMBOS} {RS}'.format(PATH=model_dir,ML=ml,DATASET=args.INPUT_FILE,SAVEFILE=save_file,N_COMBOS=args.N_COMBOS,RS=random_state)) 114 | else: 115 | all_commands.append('python {PATH}/{ML}.py {DATASET} {SAVEFILE} {RS}'.format( 116 | PATH=model_dir,ML=ml,DATASET=args.INPUT_FILE, 117 | SAVEFILE=save_file,RS=random_state) 118 | ) 119 | job_info.append({'ml':ml,'dataset':dataset,'results_path':results_path}) 120 | 121 | if args.LSF: # bsub commands 122 | for i,run_cmd in enumerate(all_commands): 123 | job_name = job_info[i]['ml'] + '_' + job_info[i]['dataset'] 124 | out_file = job_info[i]['results_path'] + job_name + '_%J.out' 125 | 126 | bsub_cmd = ('bsub -o {OUT_FILE} -n {N_CORES} -J {JOB_NAME} -q {QUEUE} ' 127 | '-R "span[hosts=1] rusage[mem={M}]" -M {M} ').format(OUT_FILE=out_file, 128 | JOB_NAME=job_name, 129 | QUEUE=args.QUEUE, 130 | N_CORES=args.N_JOBS, 131 | M=args.M) 132 | 133 | bsub_cmd += '"' + run_cmd + '"' 134 | print(bsub_cmd) 135 | os.system(bsub_cmd) # submit jobs 136 | else: # run locally 137 | Parallel(n_jobs=args.N_JOBS)(delayed(os.system)(run_cmd) for run_cmd in all_commands ) 138 | -------------------------------------------------------------------------------- /compare.py: -------------------------------------------------------------------------------- 1 | import pandas as pd 2 | import numpy as np 3 | import matplotlib as mpl 4 | mpl.rcParams['xtick.labelsize'] = 8 5 | import matplotlib.pyplot as plt 6 | import seaborn as sns 7 | sns.set_style("whitegrid") 8 | import math 9 | import argparse 10 | from glob import glob 11 | import pdb 12 | def main(): 13 | """Analyzes results and generates figures.""" 14 | 15 | parser = argparse.ArgumentParser(description="An analyst for quick ML applications.", 16 | add_help=False) 17 | 18 | parser.add_argument('RUN_DIR', action='store', type=str, help='Path to results from analysis.') 19 | parser.add_argument('-max_feat',action='store',dest='MAX_FEAT',default=10,type=int, 20 | help = 'Max features to show in importance plots.') 21 | args = parser.parse_args() 22 | 23 | # dataset = args.NAME 24 | # dataset = args.NAME.split('/')[-1].split('.')[0] 25 | # run_dir = 'results/' + dataset + '/' 26 | run_dir = args.RUN_DIR 27 | if run_dir[-1] != '/': 28 | run_dir += '/' 29 | dataset = run_dir.split('/')[-2] 30 | print('dataset:',dataset) 31 | print('loading data from',run_dir) 32 | 33 | frames = [] # data frames to combine 34 | count = 0 35 | for f in glob(run_dir + '*.csv'): 36 | if 'imp_score' not in f: 37 | frames.append(pd.read_csv(f,sep='\t',index_col=False)) 38 | count = count + 1 39 | 40 | df = pd.concat(frames, join='outer', ignore_index=True) 41 | df['prep_alg'] = df['preprocessor'] + '_' + df['algorithm'] 42 | 43 | print('loaded',count,'result files with results from these learners:',df['prep_alg'].unique()) 44 | 45 | ########################################################################## performance boxplots 46 | restricted_cols = ['prep_alg','preprocessor', 'prep-parameters', 'algorithm', 47 | 'alg-parameters','dataset', 'trial','seed','parameters'] 48 | columns_to_plot = [c for c in df.columns if c not in restricted_cols ] 49 | #['accuracy','f1_macro','bal_accuracy'] 50 | print('generating boxplots for these columns:',columns_to_plot) 51 | 52 | for col in columns_to_plot: 53 | fig = plt.figure() 54 | unique_preps = df['preprocessor'].unique() 55 | for i, prep in enumerate(unique_preps): 56 | fig.add_subplot(math.ceil(len(unique_preps)), 2,i+1) 57 | # sns.boxplot(data=df.loc[df['preprocessor']==prep],x="algorithm",y=col,orient='h') 58 | sns.boxplot(data=df.loc[df['preprocessor']==prep],y="algorithm",x=col,orient='h') 59 | plt.title(prep,size=16) 60 | # plt.gca().set_xticklabels(df.algorithm.unique(),size=10,rotation=45) 61 | # plt.ylabel(col,size=16) 62 | # plt.xlabel(col,size=16) 63 | plt.ylabel(col,size=16) 64 | # plt.ylim(0.5,1.0) 65 | # plt.xlim(0.5,1.0) 66 | # plt.xlabel('') 67 | plt.ylabel('') 68 | # plt.gca().set_xticklabels(size=10) 69 | plt.tight_layout() 70 | plt.savefig(run_dir + '_'.join([ dataset, col,'boxplots.pdf'])) 71 | 72 | ###################################################################### feature importance plots 73 | frames = [] # data frames to combine 74 | count = 0 75 | for f in glob(run_dir + '*.imp_score'): 76 | frames.append(pd.read_csv(f,sep='\t',index_col=False)) 77 | count = count + 1 78 | 79 | df = pd.concat(frames, join='outer', ignore_index=True) 80 | df['prep_alg'] = df['preprocessor'] + '_' + df['algorithm'] 81 | print('loaded',count,'feature importance files with results from these learners:', 82 | df['prep_alg'].unique()) 83 | 84 | dfp = df.groupby(['prep_alg','feature']).median().unstack(['prep_alg']) 85 | dfpn = df.groupby(['feature','prep_alg']).median().groupby('feature').sum().unstack()['score'] 86 | # print(dfpn.columns) 87 | # pdb.set_trace() 88 | dfpn = dfpn.sort_values(ascending=True) #, inplace=True) 89 | print('dfpn.head:',dfpn.tail(100)) 90 | # sort by median feature importance 91 | nf = min(args.MAX_FEAT, dfpn.index.shape[0]) 92 | 93 | dfpw = dfp.loc[dfpn.index[-nf:]] 94 | h = dfpw['score'].plot(kind='barh', stacked=True) 95 | leg = plt.legend(loc='center left', bbox_to_anchor=(1, 0.5)) 96 | fig = plt.gcf() 97 | fig.set_size_inches(5,15) 98 | plt.ylabel('Importance Score') 99 | plt.savefig(run_dir + '_'.join([ dataset, 'importance_scores.pdf']), 100 | bbox_extra_artists=(leg,h), bbox_inches='tight',figsize=(5,15)) 101 | 102 | #################################################################################### roc curves 103 | frames = [] # data frames to combine 104 | count = 0 105 | for f in glob(run_dir + '*.roc'): 106 | frames.append(pd.read_csv(f,sep='\t',index_col=False)) 107 | count = count + 1 108 | 109 | df = pd.concat(frames, join='outer', ignore_index=True) 110 | df['prep_alg'] = df['preprocessor'] + '_' + df['algorithm'] 111 | 112 | print('loaded',count,'roc files with results from these learners:',df['prep_alg'].unique()) 113 | 114 | h, ax = plt.subplots() 115 | ax.plot([0, 1],[0, 1],'--k',label='_nolegend_') 116 | # colors = ('r','y','b','g','c','k') 117 | # colors = plt.cm.Blues(np.linspace(0.1, 0.9, len(df['prep_alg'].unique()))) 118 | colors = plt.cm.Spectral(np.linspace(0.1, 0.9, len(df['prep_alg'].unique()))) 119 | n_algs = len(df['prep_alg'].unique()) 120 | # markers = ['o','v','^','<','>','8','s', 121 | # 'p','P','*','h','H','+','x','X','D','d','|','_'] 122 | for i, (alg,df_g) in enumerate(df.groupby('prep_alg')): 123 | 124 | aucs = df_g.auc.values 125 | seed_max = df_g.loc[df_g.auc.idxmax()]['seed'] 126 | seed_min = df_g.loc[df_g.auc.idxmin()]['seed'] 127 | seed_med = df_g.loc[np.abs(df_g.auc - df_g.auc.median()) == 128 | np.min(np.abs(df_g.auc - df_g.auc.median()))]['seed'] 129 | seed_med = seed_med.iloc[0] 130 | 131 | auc = df_g.auc.median() 132 | # fpr = df_g['fpr'].unique() 133 | tprs,fprs=[],[] 134 | fpr_min = df_g.loc[df_g.seed == seed_min,:]['fpr'] 135 | fpr_max = df_g.loc[df_g.seed == seed_max,:]['fpr'] 136 | tpr_min = df_g.loc[df_g.seed == seed_min,:]['tpr'] 137 | tpr_max = df_g.loc[df_g.seed == seed_max,:]['tpr'] 138 | tpr_med = df_g.loc[df_g.seed == seed_med,:]['tpr'] 139 | fpr_med = df_g.loc[df_g.seed == seed_med,:]['fpr'] 140 | 141 | ax.plot(fpr_med,tpr_med, color=colors[i % n_algs], linestyle='-', linewidth=1.5, 142 | label='{:s} (AUC = {:0.2f})'.format(alg,auc)) 143 | ax.plot(fpr_max,tpr_max, color=colors[i % n_algs], linestyle='--', linewidth=0.5, 144 | label='_nolegend_') 145 | ax.plot(fpr_min,tpr_min,color=colors[i % n_algs], linestyle='--', linewidth=0.5, 146 | label='_nolegend_'.format(alg,auc)) 147 | 148 | plt.ylabel('True Positive Rate') 149 | plt.xlabel('False Positive Rate') 150 | leg = plt.legend(loc='center left', bbox_to_anchor=(1, 0.5)) 151 | plt.ylim(0,1) 152 | plt.xlim(0,1) 153 | plt.savefig(run_dir + '_'.join([ dataset, col,'roc_curves.pdf']), bbox_inches='tight') 154 | 155 | print('done!') 156 | 157 | if __name__ == '__main__': 158 | main() 159 | -------------------------------------------------------------------------------- /metrics.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | """ 4 | Copyright 2016 Randal S. Olson 5 | 6 | This file is part of the TPOT library. 7 | 8 | The TPOT library is free software: you can redistribute it and/or 9 | modify it under the terms of the GNU General Public License as published by the 10 | Free Software Foundation, either version 3 of the License, or (at your option) 11 | any later version. 12 | 13 | The TPOT library is distributed in the hope that it will be useful, but 14 | WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 15 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more 16 | details. You should have received a copy of the GNU General Public License along 17 | with the TPOT library. If not, see http://www.gnu.org/licenses/. 18 | 19 | """ 20 | from sklearn.metrics import f1_score 21 | import numpy as np 22 | 23 | def balanced_accuracy_score(y_true, y_pred): 24 | """Default scoring function: balanced accuracy 25 | 26 | Balanced accuracy computes each class' accuracy on a per-class basis using a 27 | one-vs-rest encoding, then computes an unweighted average of the class accuracies. 28 | 29 | Parameters 30 | ---------- 31 | y_true: numpy.ndarray {n_samples} 32 | True class labels 33 | y_pred: numpy.ndarray {n_samples} 34 | Predicted class labels by the estimator 35 | 36 | Returns 37 | ------- 38 | fitness: float 39 | Returns a float value indicating the `individual`'s balanced accuracy 40 | 0.5 is as good as chance, and 1.0 is perfect predictive accuracy 41 | """ 42 | all_classes = list(set(np.append(y_true, y_pred))) 43 | all_class_accuracies = [] 44 | for this_class in all_classes: 45 | this_class_sensitivity = \ 46 | float(sum((y_pred == this_class) & (y_true == this_class))) /\ 47 | float(sum((y_true == this_class))) 48 | 49 | this_class_specificity = \ 50 | float(sum((y_pred != this_class) & (y_true != this_class))) /\ 51 | float(sum((y_true != this_class))) 52 | 53 | this_class_accuracy = (this_class_sensitivity + this_class_specificity) / 2. 54 | all_class_accuracies.append(this_class_accuracy) 55 | 56 | return np.mean(all_class_accuracies) 57 | 58 | def f1_macro(y_true,y_pred): 59 | return f1_score(y_true,y_pred,average='macro') 60 | -------------------------------------------------------------------------------- /ml/grid_search/AdaBoostClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.ensemble import AdaBoostClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, AdaBoostClassifier] 14 | pipeline_parameters = {} 15 | 16 | learning_rate_values = [0.01, 0.1, 0.5, 1.0, 10.0, 50.0, 100.0] 17 | n_estimators_values = [10, 50, 100, 500] 18 | random_state = [random_seed] 19 | 20 | all_param_combinations = itertools.product(learning_rate_values, n_estimators_values, random_state) 21 | pipeline_parameters[AdaBoostClassifier] = [{'learning_rate': learning_rate, 'n_estimators': n_estimators, 'random_state': random_state} 22 | for (learning_rate, n_estimators, random_state) in all_param_combinations] 23 | 24 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 25 | -------------------------------------------------------------------------------- /ml/grid_search/BernoulliNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import MinMaxScaler 6 | from sklearn.naive_bayes import BernoulliNB 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [MinMaxScaler, BernoulliNB] 14 | pipeline_parameters = {} 15 | 16 | alpha_values = [0., 0.1, 0.25, 0.5, 0.75, 1., 5., 10., 25., 50.] 17 | fit_prior_values = [True, False] 18 | binarize_values = [0., 0.1, 0.25, 0.5, 0.75, 0.9, 1.] 19 | pipeline_parameters[BernoulliNB] = [{'alpha': args[0], 'fit_prior': args[1], 'binarize': args[2]} 20 | for args in itertools.product(alpha_values, fit_prior_values, binarize_values)] 21 | 22 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 23 | -------------------------------------------------------------------------------- /ml/grid_search/DecisionTreeClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.tree import DecisionTreeClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, DecisionTreeClassifier] 14 | pipeline_parameters = {} 15 | 16 | min_impurity_decrease_values = np.arange(0., 0.005, 0.00025) 17 | max_features_values = [0.1, 0.25, 0.5, 0.75, 'sqrt', 'log2', None] 18 | criterion_values = ['gini', 'entropy'] 19 | random_state = [random_seed] 20 | 21 | all_param_combinations = itertools.product(min_impurity_decrease_values, max_features_values, criterion_values, random_state) 22 | pipeline_parameters[DecisionTreeClassifier] = \ 23 | [{'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'criterion': criterion, 'random_state': random_state} 24 | for (min_impurity_decrease, max_features, criterion, random_state) in all_param_combinations] 25 | 26 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 27 | -------------------------------------------------------------------------------- /ml/grid_search/ExtraTreesClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, ExtraTreesClassifier] 14 | pipeline_parameters = {} 15 | 16 | n_estimators_values = [10, 50, 100, 500] 17 | min_impurity_decrease_values = np.arange(0., 0.005, 0.00025) 18 | max_features_values = [0.1, 0.25, 0.5, 0.75, 'sqrt', 'log2', None] 19 | criterion_values = ['gini', 'entropy'] 20 | random_state = [random_seed] 21 | 22 | all_param_combinations = itertools.product(n_estimators_values, min_impurity_decrease_values, max_features_values, criterion_values, random_state) 23 | pipeline_parameters[ExtraTreesClassifier] = \ 24 | [{'n_estimators': n_estimators, 'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'criterion': criterion, 'random_state': random_state} 25 | for (n_estimators, min_impurity_decrease, max_features, criterion, random_state) in all_param_combinations] 26 | 27 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 28 | -------------------------------------------------------------------------------- /ml/grid_search/GaussianNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.naive_bayes import GaussianNB 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, GaussianNB] 14 | pipeline_parameters = {} 15 | pipeline_parameters[GaussianNB] = [{}] 16 | 17 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 18 | -------------------------------------------------------------------------------- /ml/grid_search/GradientBoostingClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.ensemble import GradientBoostingClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, GradientBoostingClassifier] 14 | pipeline_parameters = {} 15 | 16 | n_estimators_values = [10, 100, 500] 17 | min_impurity_decrease_values = np.arange(0., 0.005, 0.001) 18 | max_features_values = ['sqrt', 'log2', None] 19 | learning_rate_values = np.logspace(-2,2,5) 20 | loss_values = ['deviance', 'exponential'] 21 | random_state = [random_seed] 22 | 23 | all_param_combinations = itertools.product(n_estimators_values, min_impurity_decrease_values, max_features_values, learning_rate_values, loss_values, random_state) 24 | pipeline_parameters[GradientBoostingClassifier] = \ 25 | [{'n_estimators': n_estimators, 'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'learning_rate': learning_rate, 'loss': loss, 'random_state': random_state} 26 | for (n_estimators, min_impurity_decrease, max_features, learning_rate, loss, random_state) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 29 | -------------------------------------------------------------------------------- /ml/grid_search/KNeighborsClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.neighbors import KNeighborsClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, KNeighborsClassifier] 14 | pipeline_parameters = {} 15 | 16 | n_neighbors_values = list(range(1, 26)) + [50, 100] 17 | weights_values = ['uniform', 'distance'] 18 | 19 | all_param_combinations = itertools.product(n_neighbors_values, weights_values) 20 | pipeline_parameters[KNeighborsClassifier] = \ 21 | [{'n_neighbors': n_neighbors, 'weights': weights} 22 | for (n_neighbors, weights) in all_param_combinations] 23 | 24 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 25 | -------------------------------------------------------------------------------- /ml/grid_search/LinearSVC.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.svm import LinearSVC 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, LinearSVC] 14 | pipeline_parameters = {} 15 | 16 | C_values = np.concatenate((np.arange(0., 1.0, 0.1), np.arange(1., 10.01, 1.))) 17 | loss_values = ['hinge', 'squared_hinge'] 18 | penalty_values = ['l1', 'l2'] 19 | dual_values = [True, False] 20 | fit_intercept_values = [True, False] 21 | random_state = [random_seed] 22 | 23 | all_param_combinations = itertools.product(C_values, loss_values, penalty_values, dual_values, fit_intercept_values, random_state) 24 | pipeline_parameters[LinearSVC] = \ 25 | [{'C': C, 'penalty': penalty, 'fit_intercept': fit_intercept, 'dual': dual, 'random_state': random_state} 26 | for (C, loss, penalty, dual, fit_intercept, random_state) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 29 | -------------------------------------------------------------------------------- /ml/grid_search/LogisticRegression.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.linear_model import LogisticRegression 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, LogisticRegression] 14 | pipeline_parameters = {} 15 | 16 | C_values = np.arange(0.5, 20.1, 0.5) 17 | penalty_values = ['l1', 'l2'] 18 | fit_intercept_values = [True, False] 19 | dual_values = [True, False] 20 | random_state = [random_seed] 21 | 22 | all_param_combinations = itertools.product(C_values, penalty_values, fit_intercept_values, dual_values, random_state) 23 | pipeline_parameters[LogisticRegression] = \ 24 | [{'C': C, 'penalty': penalty, 'fit_intercept': fit_intercept, 'dual': dual, 'random_state': random_state} 25 | for (C, penalty, fit_intercept, dual, random_state) in all_param_combinations 26 | if not (penalty != 'l2' and dual != False)] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 29 | -------------------------------------------------------------------------------- /ml/grid_search/MultinomialNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import MinMaxScaler 6 | from sklearn.naive_bayes import MultinomialNB 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [MinMaxScaler, MultinomialNB] 14 | pipeline_parameters = {} 15 | 16 | alpha_values = [0., 0.1, 0.25, 0.5, 0.75, 1., 5., 10., 25., 50.] 17 | fit_prior_values = [True, False] 18 | 19 | all_param_combinations = itertools.product(alpha_values, fit_prior_values) 20 | pipeline_parameters[MultinomialNB] = \ 21 | [{'alpha': alpha, 'fit_prior': fit_prior} 22 | for (alpha, fit_prior) in all_param_combinations] 23 | 24 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 25 | -------------------------------------------------------------------------------- /ml/grid_search/PassiveAggressiveClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.linear_model import PassiveAggressiveClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, PassiveAggressiveClassifier] 14 | pipeline_parameters = {} 15 | 16 | C_values = [0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, 0.5, 1., 10., 50., 100.] 17 | loss_values = ['hinge', 'squared_hinge'] 18 | fit_intercept_values = [True, False] 19 | random_state = [random_seed] 20 | 21 | all_param_combinations = itertools.product(C_values, loss_values, fit_intercept_values, random_state) 22 | pipeline_parameters[PassiveAggressiveClassifier] = \ 23 | [{'C': C, 'loss': loss, 'fit_intercept': fit_intercept, 'random_state': random_state} 24 | for (C, loss, fit_intercept, random_state) in all_param_combinations] 25 | 26 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 27 | -------------------------------------------------------------------------------- /ml/grid_search/RandomForestClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.ensemble import RandomForestClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, RandomForestClassifier] 14 | pipeline_parameters = {} 15 | 16 | n_estimators_values = [10, 100, 500] 17 | min_impurity_decrease_values = np.arange(0., 0.005, 0.001) 18 | max_features_values = ['sqrt', 'log2', None] 19 | criterion_values = ['gini', 'entropy'] 20 | random_state = [random_seed] 21 | 22 | all_param_combinations = itertools.product(n_estimators_values, min_impurity_decrease_values, max_features_values, criterion_values, random_state) 23 | pipeline_parameters[RandomForestClassifier] = \ 24 | [{'n_estimators': n_estimators, 'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'criterion': criterion, 'random_state': random_state} 25 | for (n_estimators, min_impurity_decrease, max_features, criterion, random_state) in all_param_combinations] 26 | 27 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 28 | -------------------------------------------------------------------------------- /ml/grid_search/SGDClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.linear_model import SGDClassifier 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, SGDClassifier] 14 | pipeline_parameters = {} 15 | 16 | loss_values = ['hinge', 'log', 'modified_huber', 'squared_hinge', 'perceptron'] 17 | penalty_values = ['l2', 'l1', 'elasticnet'] 18 | alpha_values = [0.000001, 0.00001, 0.0001, 0.001, 0.01] 19 | learning_rate_values = ['constant', 'optimal', 'invscaling'] 20 | fit_intercept_values = [True, False] 21 | l1_ratio_values = [0., 0.1, 0.15, 0.25, 0.5, 0.75, 0.9, 1.] 22 | eta0_values = [0.0, 0.01, 0.1, 0.5, 1., 10., 50., 100.] 23 | power_t_values = [0., 0.1, 0.5, 1., 10., 50., 100.] 24 | random_state = [random_seed] 25 | 26 | all_param_combinations = itertools.product(loss_values, penalty_values, alpha_values, learning_rate_values, fit_intercept_values, l1_ratio_values, eta0_values, power_t_values, random_state) 27 | pipeline_parameters[SGDClassifier] = \ 28 | [{'loss': loss, 'penalty': penalty, 'alpha': alpha, 'learning_rate': learning_rate, 'fit_intercept': fit_intercept, 'l1_ratio': l1_ratio, 'eta0': eta0, 'power_t': power_t, 'random_state': random_state} 29 | for (loss, penalty, alpha, learning_rate, fit_intercept, l1_ratio, eta0, power_t, random_state) in all_param_combinations 30 | if not (penalty != 'elasticnet' and l1_ratio != 0.15) and not (learning_rate not in ['constant', 'invscaling'] and eta0 != 0.0) and not (learning_rate != 'invscaling' and power_t != 0.5)] 31 | 32 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 33 | -------------------------------------------------------------------------------- /ml/grid_search/SVC.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from sklearn.svm import SVC 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, SVC] 14 | pipeline_parameters = {} 15 | 16 | C_values = [0.01, 0.1, 0.5, 1., 10., 50., 100.] 17 | gamma_values = [0.01, 0.1, 0.5, 1., 10., 50., 100., 'auto'] 18 | kernel_values = ['poly', 'rbf', 'sigmoid'] 19 | degree_values = [2, 3] 20 | coef0_values = [0., 0.1, 0.5, 1., 10., 50., 100.] 21 | random_state = [random_seed] 22 | 23 | all_param_combinations = itertools.product(C_values, gamma_values, kernel_values, degree_values, coef0_values, random_state) 24 | pipeline_parameters[SVC] = \ 25 | [{'C': C, 'gamma': gamma, 'kernel': kernel, 'degree': degree, 'coef0': coef0, 'random_state': random_state} 26 | for (C, gamma, kernel, degree, coef0, random_state) in all_param_combinations 27 | if not (kernel != 'poly' and degree > 2) and not (kernel not in ['poly', 'sigmoid'] and coef0 != 0.0)] 28 | 29 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 30 | -------------------------------------------------------------------------------- /ml/grid_search/XGBClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | import itertools 5 | from sklearn.preprocessing import RobustScaler 6 | from xgboost import XGBClassifier # Assumes XGBoost v0.6 7 | from evaluate_model import evaluate_model 8 | 9 | dataset = sys.argv[1] 10 | save_file = sys.argv[2] 11 | random_seed = int(sys.argv[3]) 12 | 13 | pipeline_components = [RobustScaler, XGBClassifier] 14 | pipeline_parameters = {} 15 | 16 | n_estimators_values = [10, 50, 100, 500] 17 | learning_rate_values = [0.01, 0.1, 0.5, 1.0, 10.0, 50.0, 100.0] 18 | gamma_values = np.arange(0., 0.51, 0.05) 19 | max_depth_values = [1, 2, 3, 4, 5, 10, 20, 50, None] 20 | subsample_values = np.arange(0.0, 1.01, 0.1) 21 | random_state = [random_seed] 22 | 23 | all_param_combinations = itertools.product(n_estimators_values, learning_rate_values, gamma_values, max_depth_values, subsample_values, random_state) 24 | pipeline_parameters[XGBClassifier] = \ 25 | [{'n_estimators': n_estimators, 'learning_rate': learning_rate, 'gamma': gamma, 'max_depth': max_depth, 'subsample': subsample, 'seed': random_state, 'nthread': 1} 26 | for (n_estimators, learning_rate, gamma, max_depth, subsample, random_state) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_seed) 29 | -------------------------------------------------------------------------------- /ml/grid_search/evaluate_model.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import itertools 3 | import pandas as pd 4 | from sklearn.model_selection import cross_val_predict, StratifiedKFold 5 | from sklearn.metrics import accuracy_score, f1_score 6 | from sklearn.pipeline import make_pipeline 7 | from tpot_metrics import balanced_accuracy_score 8 | import warnings 9 | 10 | def evaluate_model(dataset, pipeline_components, pipeline_parameters, save_file, random_state): 11 | compression = 'gzip' if '.gz' in dataset else None 12 | input_data = pd.read_csv(dataset, compression=compression, engine='python', sep=None) 13 | features = input_data.drop('class', axis=1).values.astype(float) 14 | labels = input_data['class'].values 15 | # features, labels, feature_names = read_file(dataset, label) 16 | 17 | pipelines = [dict(zip(pipeline_parameters.keys(), list(parameter_combination))) 18 | for parameter_combination in itertools.product(*pipeline_parameters.values())] 19 | 20 | with warnings.catch_warnings(): 21 | # Squash warning messages. Turn this off when debugging! 22 | warnings.simplefilter('ignore') 23 | 24 | for pipe_parameters in pipelines: 25 | pipeline = [] 26 | for component in pipeline_components: 27 | if component in pipe_parameters: 28 | args = pipe_parameters[component] 29 | pipeline.append(component(**args)) 30 | else: 31 | pipeline.append(component()) 32 | 33 | try: 34 | clf = make_pipeline(*pipeline) 35 | cv_predictions = cross_val_predict(estimator=clf, X=features, y=labels, 36 | cv=StratifiedKFold(n_splits=10, shuffle=True, random_state=random_state)) 37 | accuracy = accuracy_score(labels, cv_predictions) 38 | macro_f1 = f1_score(labels, cv_predictions, average='macro') 39 | balanced_accuracy = balanced_accuracy_score(labels, cv_predictions) 40 | except KeyboardInterrupt: 41 | sys.exit(1) 42 | # This is a catch-all to make sure that the evaluation won't crash due to a bad parameter 43 | # combination or bad data. Turn this off when debugging! 44 | except Exception as e: 45 | continue 46 | 47 | classifier_class = pipeline_components[-1] 48 | param_string = ','.join(['{}={}'.format(parameter, value) 49 | for parameter, value in pipe_parameters[classifier_class].items()]) 50 | 51 | out_text = '\t'.join([dataset.split('/')[-1][:-7], 52 | classifier_class.__name__, 53 | param_string, 54 | str(accuracy), 55 | str(macro_f1), 56 | str(balanced_accuracy)]) 57 | 58 | print(out_text) 59 | with open(save_file, 'a') as out: 60 | out.write(out_text+'\n') 61 | 62 | sys.stdout.flush() 63 | -------------------------------------------------------------------------------- /ml/grid_search/tpot_metrics.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | """ 4 | Copyright 2016 Randal S. Olson 5 | 6 | This file is part of the TPOT library. 7 | 8 | The TPOT library is free software: you can redistribute it and/or 9 | modify it under the terms of the GNU General Public License as published by the 10 | Free Software Foundation, either version 3 of the License, or (at your option) 11 | any later version. 12 | 13 | The TPOT library is distributed in the hope that it will be useful, but 14 | WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 15 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more 16 | details. You should have received a copy of the GNU General Public License along 17 | with the TPOT library. If not, see http://www.gnu.org/licenses/. 18 | 19 | """ 20 | 21 | import numpy as np 22 | 23 | def balanced_accuracy_score(y_true, y_pred): 24 | """Default scoring function: balanced accuracy 25 | 26 | Balanced accuracy computes each class' accuracy on a per-class basis using a 27 | one-vs-rest encoding, then computes an unweighted average of the class accuracies. 28 | 29 | Parameters 30 | ---------- 31 | y_true: numpy.ndarray {n_samples} 32 | True class labels 33 | y_pred: numpy.ndarray {n_samples} 34 | Predicted class labels by the estimator 35 | 36 | Returns 37 | ------- 38 | fitness: float 39 | Returns a float value indicating the `individual`'s balanced accuracy 40 | 0.5 is as good as chance, and 1.0 is perfect predictive accuracy 41 | """ 42 | all_classes = list(set(np.append(y_true, y_pred))) 43 | all_class_accuracies = [] 44 | for this_class in all_classes: 45 | this_class_sensitivity = \ 46 | float(sum((y_pred == this_class) & (y_true == this_class))) /\ 47 | float(sum((y_true == this_class))) 48 | 49 | this_class_specificity = \ 50 | float(sum((y_pred != this_class) & (y_true != this_class))) /\ 51 | float(sum((y_true != this_class))) 52 | 53 | this_class_accuracy = (this_class_sensitivity + this_class_specificity) / 2. 54 | all_class_accuracies.append(this_class_accuracy) 55 | 56 | return np.mean(all_class_accuracies) 57 | -------------------------------------------------------------------------------- /ml/random_search/AdaBoostClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.ensemble import AdaBoostClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, AdaBoostClassifier] 15 | pipeline_parameters = {} 16 | 17 | learning_rate_values = np.random.uniform(low=1e-10, high=5., size=num_param_combinations) 18 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 19 | 20 | all_param_combinations = zip(learning_rate_values, n_estimators_values) 21 | pipeline_parameters[AdaBoostClassifier] = [{'learning_rate': learning_rate, 'n_estimators': n_estimators, 'random_state': 324089} 22 | for (learning_rate, n_estimators) in all_param_combinations] 23 | 24 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 25 | -------------------------------------------------------------------------------- /ml/random_search/BernoulliNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import MinMaxScaler 5 | from sklearn.naive_bayes import BernoulliNB 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [MinMaxScaler, BernoulliNB] 15 | pipeline_parameters = {} 16 | 17 | alpha_values = np.random.uniform(low=0., high=50., size=num_param_combinations) 18 | fit_prior_values = np.random.choice([True, False], size=num_param_combinations) 19 | binarize_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 20 | 21 | all_param_combinations = zip(alpha_values, fit_prior_values, binarize_values) 22 | pipeline_parameters[BernoulliNB] = [{'alpha': alpha, 'fit_prior': fit_prior, 'binarize': binarize} 23 | for (alpha, fit_prior, binarize) in all_param_combinations] 24 | 25 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 26 | -------------------------------------------------------------------------------- /ml/random_search/DecisionTreeClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.tree import DecisionTreeClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, DecisionTreeClassifier] 15 | pipeline_parameters = {} 16 | 17 | min_impurity_decrease_values = np.random.exponential(scale=0.01, size=num_param_combinations) 18 | max_features_values = np.random.choice(list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None], size=num_param_combinations) 19 | criterion_values = np.random.choice(['gini', 'entropy'], size=num_param_combinations) 20 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 21 | 22 | all_param_combinations = zip(min_impurity_decrease_values, max_features_values, criterion_values, max_depth_values) 23 | pipeline_parameters[DecisionTreeClassifier] = \ 24 | [{'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'criterion': criterion, 'max_depth': max_depth, 'random_state': 324089} 25 | for (min_impurity_decrease, max_features, criterion, max_depth) in all_param_combinations] 26 | 27 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 28 | -------------------------------------------------------------------------------- /ml/random_search/ExtraTreesClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.ensemble import ExtraTreesClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, ExtraTreesClassifier] 15 | pipeline_parameters = {} 16 | 17 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 18 | min_impurity_decrease_values = np.random.exponential(scale=0.01, size=num_param_combinations) 19 | max_features_values = np.random.choice(list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None], size=num_param_combinations) 20 | criterion_values = np.random.choice(['gini', 'entropy'], size=num_param_combinations) 21 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 22 | 23 | all_param_combinations = zip(n_estimators_values, min_impurity_decrease_values, max_features_values, criterion_values, max_depth_values) 24 | pipeline_parameters[ExtraTreesClassifier] = \ 25 | [{'n_estimators': n_estimators, 'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'criterion': criterion, 'max_depth': max_depth, 'random_state': 324089} 26 | for (n_estimators, min_impurity_decrease, max_features, criterion, max_depth) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 29 | -------------------------------------------------------------------------------- /ml/random_search/GaussianNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.naive_bayes import GaussianNB 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, GaussianNB] 15 | pipeline_parameters = {} 16 | pipeline_parameters[GaussianNB] = [{}] 17 | 18 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 19 | -------------------------------------------------------------------------------- /ml/random_search/GradientBoostingClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.ensemble import GradientBoostingClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, GradientBoostingClassifier] 15 | pipeline_parameters = {} 16 | 17 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 18 | min_impurity_decrease_values = np.random.exponential(scale=0.01, size=num_param_combinations) 19 | max_features_values = np.random.choice(list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None], size=num_param_combinations) 20 | learning_rate_values = np.random.uniform(low=1e-10, high=5., size=num_param_combinations) 21 | loss_values = np.random.choice(['deviance', 'exponential'], size=num_param_combinations) 22 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 23 | 24 | all_param_combinations = zip(n_estimators_values, min_impurity_decrease_values, max_features_values, learning_rate_values, loss_values, max_depth_values) 25 | pipeline_parameters[GradientBoostingClassifier] = \ 26 | [{'n_estimators': n_estimators, 'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'learning_rate': learning_rate, 'loss': loss, 'max_depth': max_depth, 'random_state': 324089} 27 | for (n_estimators, min_impurity_decrease, max_features, learning_rate, loss, max_depth) in all_param_combinations] 28 | 29 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 30 | -------------------------------------------------------------------------------- /ml/random_search/KNeighborsClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.neighbors import KNeighborsClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, KNeighborsClassifier] 15 | pipeline_parameters = {} 16 | 17 | n_neighbors_values = np.random.randint(low=1, high=100, size=num_param_combinations) 18 | weights_values = np.random.choice(['uniform', 'distance'], size=num_param_combinations) 19 | 20 | all_param_combinations = zip(n_neighbors_values, weights_values) 21 | pipeline_parameters[KNeighborsClassifier] = \ 22 | [{'n_neighbors': n_neighbors, 'weights': weights} 23 | for (n_neighbors, weights) in all_param_combinations] 24 | 25 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 26 | -------------------------------------------------------------------------------- /ml/random_search/LinearSVC.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.svm import LinearSVC 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, LinearSVC] 15 | pipeline_parameters = {} 16 | 17 | C_values = np.random.uniform(low=1e-10, high=10., size=num_param_combinations) 18 | loss_values = np.random.choice(['hinge', 'squared_hinge'], size=num_param_combinations) 19 | penalty_values = np.random.choice(['l1', 'l2'], size=num_param_combinations) 20 | dual_values = np.random.choice([True, False], size=num_param_combinations) 21 | fit_intercept_values = np.random.choice([True, False], size=num_param_combinations) 22 | 23 | all_param_combinations = zip(C_values, loss_values, penalty_values, dual_values, fit_intercept_values) 24 | pipeline_parameters[LinearSVC] = \ 25 | [{'C': C, 'penalty': penalty, 'fit_intercept': fit_intercept, 'dual': dual, 'random_state': 324089} 26 | for (C, loss, penalty, dual, fit_intercept) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 29 | -------------------------------------------------------------------------------- /ml/random_search/LogisticRegression.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.linear_model import LogisticRegression 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, LogisticRegression] 15 | pipeline_parameters = {} 16 | 17 | C_values = np.random.uniform(low=1e-10, high=10., size=num_param_combinations) 18 | penalty_values = np.random.choice(['l1', 'l2'], size=num_param_combinations) 19 | fit_intercept_values = np.random.choice([True, False], size=num_param_combinations) 20 | dual_values = np.random.choice([True, False], size=num_param_combinations) 21 | 22 | all_param_combinations = zip(C_values, penalty_values, fit_intercept_values, dual_values) 23 | pipeline_parameters[LogisticRegression] = \ 24 | [{'C': C, 'penalty': penalty, 'fit_intercept': fit_intercept, 'dual': False if penalty != 'l2' else dual, 'random_state': 324089} 25 | for (C, penalty, fit_intercept, dual) in all_param_combinations] 26 | 27 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 28 | -------------------------------------------------------------------------------- /ml/random_search/MultinomialNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import MinMaxScaler 5 | from sklearn.naive_bayes import MultinomialNB 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [MinMaxScaler, MultinomialNB] 15 | pipeline_parameters = {} 16 | 17 | alpha_values = np.random.uniform(low=0., high=10., size=num_param_combinations) 18 | fit_prior_values = np.random.choice([True, False], size=num_param_combinations) 19 | 20 | all_param_combinations = zip(alpha_values, fit_prior_values) 21 | pipeline_parameters[MultinomialNB] = \ 22 | [{'alpha': alpha, 'fit_prior': fit_prior} 23 | for (alpha, fit_prior) in all_param_combinations] 24 | 25 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 26 | -------------------------------------------------------------------------------- /ml/random_search/PassiveAggressiveClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.linear_model import PassiveAggressiveClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, PassiveAggressiveClassifier] 15 | pipeline_parameters = {} 16 | 17 | C_values = np.random.uniform(low=1e-10, high=10., size=num_param_combinations) 18 | loss_values = np.random.choice(['hinge', 'squared_hinge'], size=num_param_combinations) 19 | fit_intercept_values = np.random.choice([True, False], size=num_param_combinations) 20 | 21 | all_param_combinations = zip(C_values, loss_values, fit_intercept_values) 22 | pipeline_parameters[PassiveAggressiveClassifier] = \ 23 | [{'C': C, 'loss': loss, 'fit_intercept': fit_intercept, 'random_state': 324089} 24 | for (C, loss, fit_intercept) in all_param_combinations] 25 | 26 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 27 | -------------------------------------------------------------------------------- /ml/random_search/RandomForestClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.ensemble import RandomForestClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, RandomForestClassifier] 15 | pipeline_parameters = {} 16 | 17 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 18 | min_impurity_decrease_values = np.random.exponential(scale=0.01, size=num_param_combinations) 19 | max_features_values = np.random.choice(list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None], size=num_param_combinations) 20 | criterion_values = np.random.choice(['gini', 'entropy'], size=num_param_combinations) 21 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 22 | 23 | all_param_combinations = zip(n_estimators_values, min_impurity_decrease_values, max_features_values, criterion_values, max_depth_values) 24 | pipeline_parameters[RandomForestClassifier] = \ 25 | [{'n_estimators': n_estimators, 'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'criterion': criterion, 'max_depth': max_depth, 'random_state': 324089} 26 | for (n_estimators, min_impurity_decrease, max_features, criterion, max_depth) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 29 | -------------------------------------------------------------------------------- /ml/random_search/SGDClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.linear_model import SGDClassifier 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, SGDClassifier] 15 | pipeline_parameters = {} 16 | 17 | loss_values = np.random.choice(['hinge', 'log', 'modified_huber', 'squared_hinge', 'perceptron'], size=num_param_combinations) 18 | penalty_values = np.random.choice(['l2', 'l1', 'elasticnet'], size=num_param_combinations) 19 | alpha_values = np.random.exponential(scale=0.01, size=num_param_combinations) 20 | learning_rate_values = np.random.choice(['constant', 'optimal', 'invscaling'], size=num_param_combinations) 21 | fit_intercept_values = np.random.choice([True, False], size=num_param_combinations) 22 | l1_ratio_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 23 | eta0_values = np.random.uniform(low=0., high=5., size=num_param_combinations) 24 | power_t_values = np.random.uniform(low=0., high=5., size=num_param_combinations) 25 | 26 | all_param_combinations = zip(loss_values, penalty_values, alpha_values, learning_rate_values, fit_intercept_values, l1_ratio_values, eta0_values, power_t_values) 27 | pipeline_parameters[SGDClassifier] = \ 28 | [{'loss': loss, 'penalty': penalty, 'alpha': alpha, 'learning_rate': learning_rate, 'fit_intercept': fit_intercept, 29 | 'l1_ratio': 0.15 if penalty != 'elasticnet' else l1_ratio, 'eta0': 0. if learning_rate not in ['constant', 'invscaling'] else eta0, 30 | 'power_t': 0.5 if learning_rate != 'invscaling' else power_t, 'random_state': 324089} 31 | for (loss, penalty, alpha, learning_rate, fit_intercept, l1_ratio, eta0, power_t) in all_param_combinations] 32 | 33 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 34 | -------------------------------------------------------------------------------- /ml/random_search/SVC.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from sklearn.svm import SVC 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, SVC] 15 | pipeline_parameters = {} 16 | 17 | C_values = np.random.uniform(low=1e-10, high=500., size=num_param_combinations) 18 | gamma_values = np.random.choice(list(np.arange(0.05, 1.01, 0.05)) + ['auto'], size=num_param_combinations) 19 | kernel_values = np.random.choice(['poly', 'rbf', 'sigmoid'], size=num_param_combinations) 20 | degree_values = np.random.choice([2, 3], size=num_param_combinations) 21 | coef0_values = np.random.uniform(low=0., high=10., size=num_param_combinations) 22 | 23 | all_param_combinations = zip(C_values, gamma_values, kernel_values, degree_values, coef0_values) 24 | pipeline_parameters[SVC] = \ 25 | [{'C': C, 'gamma': float(gamma) if gamma != 'auto' else gamma, 'kernel': str(kernel), 'degree': 2 if kernel != 'poly' else degree, 'coef0': 0. if kernel not in ['poly', 'sigmoid'] else coef0, 'random_state': 324089} 26 | for (C, gamma, kernel, degree, coef0) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 29 | -------------------------------------------------------------------------------- /ml/random_search/XGBClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | from sklearn.preprocessing import RobustScaler 5 | from xgboost import XGBClassifier # Assumes XGBoost v0.6 6 | from evaluate_model import evaluate_model 7 | 8 | dataset = sys.argv[1] 9 | num_param_combinations = int(sys.argv[2]) 10 | random_seed = int(sys.argv[3]) 11 | 12 | np.random.seed(random_seed) 13 | 14 | pipeline_components = [RobustScaler, XGBClassifier] 15 | pipeline_parameters = {} 16 | 17 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 18 | learning_rate_values = np.random.uniform(low=1e-10, high=5., size=num_param_combinations) 19 | gamma_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 20 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 21 | subsample_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 22 | 23 | all_param_combinations = zip(n_estimators_values, learning_rate_values, gamma_values, max_depth_values, subsample_values) 24 | pipeline_parameters[XGBClassifier] = \ 25 | [{'n_estimators': n_estimators, 'learning_rate': learning_rate, 'gamma': gamma, 'max_depth': max_depth, 'subsample': subsample, 'seed': 324089, 'nthread': 1} 26 | for (n_estimators, learning_rate, gamma, max_depth, subsample) in all_param_combinations] 27 | 28 | evaluate_model(dataset, pipeline_components, pipeline_parameters) 29 | -------------------------------------------------------------------------------- /ml/random_search/evaluate_model.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import itertools 3 | import pandas as pd 4 | from sklearn.model_selection import cross_val_predict, StratifiedKFold 5 | from sklearn.metrics import accuracy_score, f1_score 6 | from sklearn.pipeline import make_pipeline 7 | from tpot_metrics import balanced_accuracy_score 8 | import warnings 9 | 10 | def evaluate_model(dataset, out_file, pipeline_components, pipeline_parameters): 11 | input_data = pd.read_csv(dataset, compression='gzip', sep='\t') 12 | features = input_data.drop('class', axis=1).values.astype(float) 13 | labels = input_data['class'].values 14 | 15 | pipelines = [dict(zip(pipeline_parameters.keys(), list(parameter_combination))) 16 | for parameter_combination in itertools.product(*pipeline_parameters.values())] 17 | 18 | with warnings.catch_warnings(): 19 | # Squash warning messages. Turn this off when debugging! 20 | warnings.simplefilter('ignore') 21 | 22 | for pipe_parameters in pipelines: 23 | pipeline = [] 24 | for component in pipeline_components: 25 | if component in pipe_parameters: 26 | args = pipe_parameters[component] 27 | pipeline.append(component(**args)) 28 | else: 29 | pipeline.append(component()) 30 | 31 | try: 32 | clf = make_pipeline(*pipeline) 33 | cv_predictions = cross_val_predict(estimator=clf, X=features, y=labels, cv=StratifiedKFold(n_splits=10, shuffle=True, random_state=90483257)) 34 | accuracy = accuracy_score(labels, cv_predictions) 35 | macro_f1 = f1_score(labels, cv_predictions, average='macro') 36 | balanced_accuracy = balanced_accuracy_score(labels, cv_predictions) 37 | except KeyboardInterrupt: 38 | sys.exit(1) 39 | # This is a catch-all to make sure that the evaluation won't crash due to a bad parameter 40 | # combination or bad data. Turn this off when debugging! 41 | except Exception as e: 42 | continue 43 | 44 | classifier_class = pipeline_components[-1] 45 | param_string = ','.join(['{}={}'.format(parameter, value) 46 | for parameter, value in pipe_parameters[classifier_class].items()]) 47 | 48 | out_text = '\t'.join([dataset.split('/')[-1][:-7], 49 | classifier_class.__name__, 50 | param_string, 51 | str(accuracy), 52 | str(macro_f1), 53 | str(balanced_accuracy)]) 54 | 55 | print(out_text) 56 | sys.stdout.flush() 57 | -------------------------------------------------------------------------------- /ml/random_search/tpot_metrics.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | """ 4 | Copyright 2016 Randal S. Olson 5 | 6 | This file is part of the TPOT library. 7 | 8 | The TPOT library is free software: you can redistribute it and/or 9 | modify it under the terms of the GNU General Public License as published by the 10 | Free Software Foundation, either version 3 of the License, or (at your option) 11 | any later version. 12 | 13 | The TPOT library is distributed in the hope that it will be useful, but 14 | WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 15 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more 16 | details. You should have received a copy of the GNU General Public License along 17 | with the TPOT library. If not, see http://www.gnu.org/licenses/. 18 | 19 | """ 20 | 21 | import numpy as np 22 | 23 | def balanced_accuracy_score(y_true, y_pred): 24 | """Default scoring function: balanced accuracy 25 | 26 | Balanced accuracy computes each class' accuracy on a per-class basis using a 27 | one-vs-rest encoding, then computes an unweighted average of the class accuracies. 28 | 29 | Parameters 30 | ---------- 31 | y_true: numpy.ndarray {n_samples} 32 | True class labels 33 | y_pred: numpy.ndarray {n_samples} 34 | Predicted class labels by the estimator 35 | 36 | Returns 37 | ------- 38 | fitness: float 39 | Returns a float value indicating the `individual`'s balanced accuracy 40 | 0.5 is as good as chance, and 1.0 is perfect predictive accuracy 41 | """ 42 | all_classes = list(set(np.append(y_true, y_pred))) 43 | all_class_accuracies = [] 44 | for this_class in all_classes: 45 | this_class_sensitivity = \ 46 | float(sum((y_pred == this_class) & (y_true == this_class))) /\ 47 | float(sum((y_true == this_class))) 48 | 49 | this_class_specificity = \ 50 | float(sum((y_pred != this_class) & (y_true != this_class))) /\ 51 | float(sum((y_true != this_class))) 52 | 53 | this_class_accuracy = (this_class_sensitivity + this_class_specificity) / 2. 54 | all_class_accuracies.append(this_class_accuracy) 55 | 56 | return np.mean(all_class_accuracies) 57 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/AdaBoostClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.ensemble import AdaBoostClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | # inputs 13 | dataset = sys.argv[1] 14 | save_file = sys.argv[2] 15 | num_param_combinations = int(sys.argv[3]) 16 | random_seed = int(sys.argv[4]) 17 | preps = sys.argv[5] 18 | label = sys.argv[6] 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | label = sys.argv[6] 26 | pipeline_components.append((p, preprocessor_dict[p])) 27 | # if pipeline_components[-1] is SelectFromModel: 28 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 29 | # elif pipeline_components[-1] is RFE: 30 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 31 | 32 | 33 | pipeline_components.append(('AdaBoostClassifier',AdaBoostClassifier())) 34 | 35 | # parameters for method 36 | learning_rate_values = np.random.uniform(low=1e-10, high=5., size=num_param_combinations) 37 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 38 | 39 | pipeline_parameters['AdaBoostClassifier'] = {'learning_rate': learning_rate_values, 'n_estimators': n_estimators_values, 'random_state': 324089} 40 | ] 41 | 42 | 43 | #evaluate 44 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label, label) 45 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/BernoulliNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.naive_bayes import BernoulliNB 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | label = sys.argv[6] 19 | 20 | np.random.seed(random_seed) 21 | 22 | # construct pipeline 23 | pipeline_components=[] 24 | pipeline_parameters={} 25 | for p in preps.split(','): 26 | pipeline_components.append((p, preprocessor_dict[p])) 27 | # if pipeline_components[-1] is SelectFromModel: 28 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 29 | # elif pipeline_components[-1] is RFE: 30 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 31 | 32 | 33 | pipeline_components.append('BernoulliNB', BernoulliNB ()) 34 | 35 | 36 | alpha_values = np.random.uniform(low=0., high=50., size=num_param_combinations) 37 | fit_prior_values = np.random.choice([True, False], size=num_param_combinations) 38 | binarize_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 39 | 40 | pipeline_parameters[BernoulliNB] = {'alpha': alpha_values, 'fit_prior': fit_prior_values, 'binarize': binarize_values} 41 | 42 | 43 | 44 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label, label) 45 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/DecisionTreeClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.tree import DecisionTreeClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | label = sys.argv[6] 19 | 20 | np.random.seed(random_seed) 21 | 22 | # construct pipeline 23 | pipeline_components=[] 24 | pipeline_parameters={} 25 | for p in preps.split(','): 26 | pipeline_components.append((p, preprocessor_dict[p])) 27 | # if pipeline_components[-1] is SelectFromModel: 28 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 29 | # elif pipeline_components[-1] is RFE: 30 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 31 | 32 | 33 | pipeline_components.append('DecisionTreeClassifier', DecisionTreeClassifier()) 34 | 35 | 36 | min_impurity_decrease_values = np.random.exponential(scale=0.01, size=num_param_combinations) 37 | max_features_values = np.random.choice(list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None], size=num_param_combinations) 38 | criterion_values = np.random.choice(['gini', 'entropy'], size=num_param_combinations) 39 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 40 | 41 | pipeline_parameters[DecisionTreeClassifier] = \ 42 | {'min_impurity_decrease': min_impurity_decrease_values, 'max_features': max_features_values, 'criterion': criterion_values, 'max_depth': max_depth_values, 'random_state': 324089} 43 | 44 | 45 | 46 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label, label) 47 | 48 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/ExtraTreesClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | 7 | from sklearn.ensemble import ExtraTreesClassifier 8 | from evaluate_model import evaluate_model 9 | from preprocessors import preprocessor_dict 10 | 11 | dataset = sys.argv[1] 12 | save_file = sys.argv[2] 13 | num_param_combinations = int(sys.argv[3]) 14 | random_seed = int(sys.argv[4]) 15 | preps = sys.argv[5] 16 | label = sys.argv[6] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('ExtraTreesClassifier', ExtraTreesClassifier ()) 33 | 34 | 35 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 36 | min_impurity_decrease_values = np.random.exponential(scale=0.01, size=num_param_combinations) 37 | max_features_values = np.random.choice(list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None], size=num_param_combinations) 38 | criterion_values = np.random.choice(['gini', 'entropy'], size=num_param_combinations) 39 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 40 | 41 | pipeline_parameters['ExtraTreesClassifier'] = \ 42 | {'n_estimators': n_estimators_values, 'min_impurity_decrease': min_impurity_decrease_values, 'max_features': max_features_values, 'criterion': criterion_values, 'max_depth': max_depth_values, 'random_state': 324089} 43 | 44 | 45 | 46 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label, label) 47 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/GaussianNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.naive_bayes import GaussianNB 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('GaussianNB', GaussianNB ()) 33 | 34 | pipeline_parameters['GaussianNB'] = {} 35 | 36 | 37 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 38 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/GradientBoostingClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.ensemble import GradientBoostingClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('GradientBoostingClassifier', GradientBoostingClassifier()) 33 | 34 | 35 | n_estimators_values = np.random.choice(list(range(50, 1001, 50)), size=num_param_combinations) 36 | min_impurity_decrease_values = np.random.exponential(scale=0.01, size=num_param_combinations) 37 | max_features_values = np.random.choice(list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None], size=num_param_combinations) 38 | learning_rate_values = np.random.uniform(low=1e-10, high=5., size=num_param_combinations) 39 | loss_values = np.random.choice(['deviance', 'exponential'], size=num_param_combinations) 40 | max_depth_values = np.random.choice(list(range(1, 51)) + [None], size=num_param_combinations) 41 | 42 | pipeline_parameters['GradientBoostingClassifier'] = \ 43 | {'n_estimators': n_estimators_values, 'min_impurity_decrease': min_impurity_decrease_values, 'max_features': max_features_values, 'learning_rate': learning_rate_values, 'loss': loss_values, 'max_depth': max_depth_values, 'random_state': 324089} 44 | 45 | 46 | 47 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 48 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/KNeighborsClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.neighbors import KNeighborsClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('KNeighborsClassifier',KNeighborsClassifier()) 33 | 34 | 35 | n_neighbors_values = np.random.randint(low=1, high=100, size=num_param_combinations) 36 | weights_values = np.random.choice(['uniform', 'distance'], size=num_param_combinations) 37 | 38 | pipeline_parameters['KNeighborsClassifier'] = \ 39 | {'n_neighbors': n_neighbors_values, 'weights': weights_values} 40 | 41 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 42 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/LinearSVC.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.svm import LinearSVC 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('LinearSVC', LinearSVC ()) 33 | 34 | 35 | C_values = np.random.uniform(low=1e-10, high=10., size=num_param_combinations) 36 | loss_values = np.random.choice(['hinge', 'squared_hinge'], size=num_param_combinations) 37 | penalty_values = np.random.choice(['l1', 'l2'], size=num_param_combinations) 38 | dual_values = np.random.choice([True, False], size=num_param_combinations) 39 | fit_intercept_values = np.random.choice([True, False], size=num_param_combinations) 40 | 41 | pipeline_parameters['LinearSVC'] = \ 42 | {'C': C_values, 'penalty': penalty_values, 'fit_intercept': fit_intercept_values, 'dual': dual_values, 'random_state': 324089} 43 | 44 | 45 | 46 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 47 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/LogisticRegression.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.linear_model import LogisticRegression 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append(('LogisticRegression', LogisticRegression(random_state=random_seed))) 33 | 34 | 35 | C_values = np.random.uniform(low=1e-10, high=10., size=num_param_combinations) 36 | penalty_values = ['l1', 'l2'] 37 | fit_intercept_values = [True, False] 38 | # dual_values = np.random.choice([True, False], size=num_param_combinations) 39 | 40 | pipeline_parameters['LogisticRegression'] = \ 41 | {'C': C_values, 'penalty': penalty_values, 'fit_intercept': fit_intercept_values} 42 | 43 | 44 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 45 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/MLPClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.neural_network import MLPClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | # inputs 13 | dataset = sys.argv[1] 14 | save_file = sys.argv[2] 15 | num_param_combinations = int(sys.argv[3]) 16 | random_seed = int(sys.argv[4]) 17 | preps = sys.argv[5] 18 | label = sys.argv[6] 19 | 20 | 21 | np.random.seed(random_seed) 22 | 23 | # construct pipeline 24 | pipeline_components = [] 25 | pipeline_parameters = {} 26 | 27 | for p in preps.split(','): 28 | pipeline_components.append((p, preprocessor_dict[p])) 29 | # if p is 'SelectFromModel': 30 | # pipeline_parameters[p] = [{'estimator': }] 31 | # elif p is 'RFE': 32 | # pipeline_parameters[p] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 33 | 34 | pipeline_components.append(('MLPClassifier',MLPClassifier())) 35 | 36 | # parameters for method 37 | hidden_layer_sizes = [(n_layers,n_nodes) for n_layers in np.arange(1,10) for n_nodes in np.arange(10,100,10)] 38 | print(hidden_layer_sizes) 39 | activation = ['identity','logistic','tanh','relu'] 40 | solver = ['lbfgs', 'sgd', 'adam'] 41 | 42 | pipeline_parameters['MLPClassifier'] = \ 43 | {'hidden_layer_sizes':hidden_layer_sizes, 'activation':activation, 'solver':solver} 44 | 45 | #evaluate 46 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 47 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/MultinomialNB.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.naive_bayes import MultinomialNB 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('MultinomialNB',MultinomialNB() ) 33 | 34 | 35 | alpha_values = np.random.uniform(low=0., high=10., size=num_param_combinations) 36 | fit_prior_values = np.random.choice([True, False], size=num_param_combinations) 37 | 38 | pipeline_parameters['MultinomialNB'] = \ 39 | {'alpha': alpha_values, 'fit_prior': fit_prior_values} 40 | 41 | 42 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 43 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/PassiveAggressiveClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.linear_model import PassiveAggressiveClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('PassiveAggressiveClassifier', PassiveAggressiveClassifier()) 33 | 34 | 35 | C_values = np.random.uniform(low=1e-10, high=10., size=num_param_combinations) 36 | loss_values = np.random.choice(['hinge', 'squared_hinge'], size=num_param_combinations) 37 | fit_intercept_values = np.random.choice([True, False], size=num_param_combinations) 38 | 39 | pipeline_parameters[PassiveAggressiveClassifier] = \ 40 | {'C': C_values, 'loss': loss_values, 'fit_intercept': fit_intercept, 'random_state': 324089} 41 | 42 | 43 | 44 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 45 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/RandomForestClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.ensemble import RandomForestClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | # inputs 13 | dataset = sys.argv[1] 14 | save_file = sys.argv[2] 15 | num_param_combinations = int(sys.argv[3]) 16 | random_seed = int(sys.argv[4]) 17 | preps = sys.argv[5] 18 | label = sys.argv[6] 19 | 20 | 21 | np.random.seed(random_seed) 22 | 23 | # construct pipeline 24 | pipeline_components = [] 25 | pipeline_parameters = {} 26 | 27 | for p in preps.split(','): 28 | pipeline_components.append((p, preprocessor_dict[p])) 29 | # if p is 'SelectFromModel': 30 | # pipeline_parameters[p] = [{'estimator': }] 31 | # elif p is 'RFE': 32 | # pipeline_parameters[p] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 33 | 34 | pipeline_components.append(('RandomForestClassifier',RandomForestClassifier(class_weight='balanced'))) 35 | 36 | # parameters for method 37 | n_estimators= list(range(50, 1001, 50)) 38 | min_impurity_decrease= np.random.exponential(scale=0.01, size=num_param_combinations) 39 | max_features= list(np.arange(0.01, 1., 0.01)) + ['sqrt', 'log2', None] 40 | criterion= ['gini', 'entropy'] 41 | max_depth= list(range(1, 21)) + [None] 42 | 43 | pipeline_parameters['RandomForestClassifier'] = \ 44 | {'n_estimators': n_estimators, 'min_impurity_decrease': min_impurity_decrease, 'max_features': max_features, 'criterion': criterion, 'max_depth': max_depth, 'random_state': [324089], 'class_weight':['balanced']} 45 | 46 | #evaluate 47 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 48 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/SGDClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.linear_model import SGDClassifier 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append('SGDClassifier', SGDClassifier()) 33 | 34 | 35 | loss_values = np.random.choice(['hinge', 'log', 'modified_huber', 'squared_hinge', 'perceptron'], size=num_param_combinations) 36 | penalty_values = np.random.choice(['l2', 'l1', 'elasticnet'], size=num_param_combinations) 37 | alpha_values = np.random.exponential(scale=0.01, size=num_param_combinations) 38 | learning_rate_values = np.random.choice(['constant', 'optimal', 'invscaling'], size=num_param_combinations) 39 | fit_intercept_values = np.random.choice([True, False], size=num_param_combinations) 40 | l1_ratio_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 41 | eta0_values = np.random.uniform(low=0., high=5., size=num_param_combinations) 42 | power_t_values = np.random.uniform(low=0., high=5., size=num_param_combinations) 43 | 44 | pipeline_parameters['SGDClassifier'] = \ 45 | [{'loss': loss_values, 'penalty': penalty, 'alpha': alpha_values, 'learning_rate': learning_rate, 'fit_intercept': fit_intercept_values, 46 | 'l1_ratio': 0.15 if penalty != 'elasticnet' else l1_ratio, 'eta0': 0. if learning_rate not in ['constant', 'invscaling'] else eta0, 47 | 'power_t': 0.5 if learning_rate != 'invscaling' else power_t, 'random_state': 324089} 48 | for penalty, l1_ratio, learning_rate, eta0, power_t in zip(penalty_values, l1_ratio_values, learning_rate_values, eta0_values, power_t_values)] 49 | 50 | 51 | 52 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 53 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/SVC.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from sklearn.feature_selection import SelectFromModel, RFE 6 | from sklearn.ensemble import ExtraTreesClassifier 7 | 8 | from sklearn.svm import SVC 9 | from evaluate_model import evaluate_model 10 | from preprocessors import preprocessor_dict 11 | 12 | dataset = sys.argv[1] 13 | save_file = sys.argv[2] 14 | num_param_combinations = int(sys.argv[3]) 15 | random_seed = int(sys.argv[4]) 16 | preps = sys.argv[5] 17 | label = sys.argv[6] 18 | 19 | np.random.seed(random_seed) 20 | 21 | # construct pipeline 22 | pipeline_components=[] 23 | pipeline_parameters={} 24 | for p in preps.split(','): 25 | pipeline_components.append((p, preprocessor_dict[p])) 26 | # if pipeline_components[-1] is SelectFromModel: 27 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | # elif pipeline_components[-1] is RFE: 29 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 30 | 31 | 32 | pipeline_components.append(('SVC', SVC(random_state=random_seed))) 33 | 34 | 35 | C_values = np.random.uniform(low=1e-10, high=500., size=num_param_combinations) 36 | gamma_values = np.random.choice(list(np.arange(0.05, 1.01, 0.05)) + ['auto'], size=num_param_combinations) 37 | kernel_values = ['poly', 'rbf', 'sigmoid'] 38 | degree_values = [2, 3] 39 | coef0_values = np.random.uniform(low=0., high=10., size=num_param_combinations) 40 | 41 | pipeline_parameters['SVC'] = \ 42 | {'C': C_values, 'gamma': [float(gamma) if gamma != 'auto' else gamma for gamma in gamma_values], 'kernel': kernel_values, 43 | 'degree': degree_values, 'coef0': coef0_values} 44 | 45 | 46 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 47 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/XGBClassifier.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import pandas as pd 3 | import numpy as np 4 | 5 | from xgboost import XGBClassifier # Assumes XGBoost v0.6 6 | from evaluate_model import evaluate_model 7 | from preprocessors import preprocessor_dict 8 | from read_file import read_file 9 | 10 | dataset = sys.argv[1] 11 | save_file = sys.argv[2] 12 | num_param_combinations = int(sys.argv[3]) 13 | random_seed = int(sys.argv[4]) 14 | preps = sys.argv[5] 15 | label = sys.argv[6] 16 | 17 | np.random.seed(random_seed) 18 | 19 | # construct pipeline 20 | pipeline_components=[] 21 | pipeline_parameters={} 22 | for p in preps.split(','): 23 | pipeline_components.append((p, preprocessor_dict[p])) 24 | # if pipeline_components[-1] is SelectFromModel: 25 | # pipeline_parameters[SelectFromModel] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 26 | # elif pipeline_components[-1] is RFE: 27 | # pipeline_parameters[RFE] = [{'estimator': ExtraTreesClassifier(n_estimators=100, random_state=324089)}] 28 | 29 | # need to load data to set balanced weights.. 30 | features, labels, _ = read_file(dataset) 31 | if len(np.unique(labels))==2: 32 | frac = float(sum(labels==0))/float(sum(labels==1)) 33 | else: 34 | frac = 1 35 | 36 | pipeline_components.append(('XGBClassifier', XGBClassifier(random_seed=random_seed, n_thread=1,scale_pos_weight=frac))) 37 | 38 | 39 | n_estimators_values = list(range(50, 1001, 50)) 40 | learning_rate_values = np.random.uniform(low=1e-10, high=5., size=num_param_combinations) 41 | gamma_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 42 | max_depth_values = list(range(1, 21)) 43 | subsample_values = np.random.uniform(low=0., high=1., size=num_param_combinations) 44 | 45 | pipeline_parameters['XGBClassifier'] = \ 46 | {'n_estimators': n_estimators_values, 'learning_rate': learning_rate_values, 'gamma': gamma_values, 'max_depth': max_depth_values, 'subsample': subsample_values} 47 | 48 | 49 | 50 | evaluate_model(dataset, save_file, random_seed, pipeline_components, pipeline_parameters, num_param_combinations, label) 51 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/evaluate_model.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import itertools 3 | import pandas as pd 4 | from sklearn.model_selection import RandomizedSearchCV, StratifiedKFold, cross_val_predict 5 | from sklearn.metrics import accuracy_score, f1_score, roc_auc_score 6 | from sklearn.pipeline import Pipeline 7 | from metrics import balanced_accuracy_score 8 | import warnings 9 | 10 | from tempfile import mkdtemp 11 | from shutil import rmtree 12 | from sklearn.externals.joblib import Memory 13 | from read_file import read_file 14 | from utils import feature_importance , roc 15 | import pdb 16 | 17 | def evaluate_model(dataset, save_file, random_state, pipeline_components, pipeline_parameters, n_combos, label): 18 | 19 | features, labels, feature_names = read_file(dataset, label) 20 | # pipelines = [dict(zip(pipeline_parameters.keys(), list(parameter_combination))) 21 | # for parameter_combination in itertools.product(*pipeline_parameters.values())] 22 | 23 | # Create a temporary folder to store the transformers of the pipeline 24 | cachedir = mkdtemp() 25 | memory = Memory(cachedir=cachedir, verbose=0) 26 | 27 | # print ( pipeline_components) 28 | # print(pipeline_parameters) 29 | with warnings.catch_warnings(): 30 | # Squash warning messages. Turn this off when debugging! 31 | warnings.simplefilter('ignore') 32 | cv = StratifiedKFold(n_splits=10, shuffle=True,random_state=random_state) 33 | hyperparameters = {} 34 | for k,v in pipeline_parameters.items(): 35 | for param,pvals in v.items(): 36 | hyperparameters.update({k+'__'+param:pvals}) 37 | pipeline = Pipeline(pipeline_components, memory=memory) 38 | 39 | # run Randomized Search CV to tune the hyperparameter settings 40 | est = RandomizedSearchCV(estimator=pipeline, param_distributions = hyperparameters, n_iter = n_combos, 41 | cv=cv, random_state=random_state, refit=True, 42 | error_score=0.0) 43 | est.fit(features, labels) 44 | best_est = est.best_estimator_ 45 | # generate cross-validated predictions for each data point using the best estimator 46 | cv_predictions = cross_val_predict(estimator=best_est, X=features, y=labels, cv=cv) 47 | 48 | # get cv probabilities 49 | skip = False 50 | if getattr(best_est, "predict_proba", None): 51 | method = "predict_proba" 52 | elif getattr(best_est, "decision_function", None): 53 | method = "decision_function" 54 | else: 55 | skip = True 56 | 57 | if not skip: 58 | cv_probabilities = cross_val_predict(estimator=best_est, X=features, y=labels, method=method, cv=cv) 59 | if method == "predict_proba": 60 | cv_probabilities = cv_probabilities[:,1] 61 | 62 | accuracy = accuracy_score(labels, cv_predictions) 63 | macro_f1 = f1_score(labels, cv_predictions, average='macro') 64 | balanced_accuracy = balanced_accuracy_score(labels, cv_predictions) 65 | roc_auc = roc_auc_score(labels,cv_probabilities) 66 | 67 | preprocessor_classes = [p[0] for p in pipeline_components[:-1]] 68 | 69 | preprocessor_param_string = 'default' 70 | for preprocessor_class in preprocessor_classes: 71 | if preprocessor_class in pipeline_parameters.keys(): 72 | preprocessor_param_string = ','.join(['{}={}'.format(parameter, '|'.join([x.strip() for x in str(value).split(',')])) 73 | for parameter, value in pipeline_parameters[preprocessor_class].items()]) 74 | 75 | classifier_class = pipeline_components[-1][0] 76 | param_string = ','.join(['{}={}'.format(p, v) for p,v in est.best_params_.items()]) 77 | # for parameter, value in pipeline_parameters[classifier_class].items()]) 78 | 79 | out_text = '\t'.join([dataset.split('/')[-1].split('.')[0], 80 | ','.join(preprocessor_classes), 81 | preprocessor_param_string, 82 | classifier_class, 83 | param_string, 84 | str(random_state), 85 | str(accuracy), 86 | str(macro_f1), 87 | str(balanced_accuracy), 88 | str(roc_auc)]) 89 | print(out_text) 90 | with open(save_file, 'a') as out: 91 | out.write(out_text+'\n') 92 | sys.stdout.flush() 93 | 94 | # write feature importances 95 | est_name = classifier_class 96 | feature_importance(save_file, best_est, est_name, feature_names, features, labels, random_state, ','.join(preprocessor_classes), preprocessor_param_string,classifier_class, param_string) 97 | # write roc curves 98 | if not skip: 99 | roc(save_file, best_est, labels, cv_probabilities, random_state, ','.join(preprocessor_classes), preprocessor_param_string,classifier_class, param_string) 100 | # Delete the temporary cache before exiting 101 | rmtree(cachedir) 102 | -------------------------------------------------------------------------------- /ml/random_search_preprocessing/tpot_metrics.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | """ 4 | Copyright 2016 Randal S. Olson 5 | 6 | This file is part of the TPOT library. 7 | 8 | The TPOT library is free software: you can redistribute it and/or 9 | modify it under the terms of the GNU General Public License as published by the 10 | Free Software Foundation, either version 3 of the License, or (at your option) 11 | any later version. 12 | 13 | The TPOT library is distributed in the hope that it will be useful, but 14 | WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 15 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more 16 | details. You should have received a copy of the GNU General Public License along 17 | with the TPOT library. If not, see http://www.gnu.org/licenses/. 18 | 19 | """ 20 | 21 | import numpy as np 22 | 23 | def balanced_accuracy_score(y_true, y_pred): 24 | """Default scoring function: balanced accuracy 25 | 26 | Balanced accuracy computes each class' accuracy on a per-class basis using a 27 | one-vs-rest encoding, then computes an unweighted average of the class accuracies. 28 | 29 | Parameters 30 | ---------- 31 | y_true: numpy.ndarray {n_samples} 32 | True class labels 33 | y_pred: numpy.ndarray {n_samples} 34 | Predicted class labels by the estimator 35 | 36 | Returns 37 | ------- 38 | fitness: float 39 | Returns a float value indicating the `individual`'s balanced accuracy 40 | 0.5 is as good as chance, and 1.0 is perfect predictive accuracy 41 | """ 42 | all_classes = list(set(np.append(y_true, y_pred))) 43 | all_class_accuracies = [] 44 | for this_class in all_classes: 45 | this_class_sensitivity = \ 46 | float(sum((y_pred == this_class) & (y_true == this_class))) /\ 47 | float(sum((y_true == this_class))) 48 | 49 | this_class_specificity = \ 50 | float(sum((y_pred != this_class) & (y_true != this_class))) /\ 51 | float(sum((y_true != this_class))) 52 | 53 | this_class_accuracy = (this_class_sensitivity + this_class_specificity) / 2. 54 | all_class_accuracies.append(this_class_accuracy) 55 | 56 | return np.mean(all_class_accuracies) 57 | -------------------------------------------------------------------------------- /preprocessors.py: -------------------------------------------------------------------------------- 1 | from sklearn.preprocessing import Binarizer, MaxAbsScaler, MinMaxScaler 2 | from sklearn.preprocessing import Normalizer, PolynomialFeatures, RobustScaler, StandardScaler 3 | from sklearn.decomposition import FastICA, PCA 4 | from sklearn.kernel_approximation import RBFSampler, Nystroem 5 | from sklearn.cluster import FeatureAgglomeration 6 | from sklearn.feature_selection import SelectFwe, SelectPercentile, VarianceThreshold 7 | from sklearn.feature_selection import SelectFromModel, RFE 8 | from sklearn.ensemble import ExtraTreesClassifier 9 | 10 | preprocessor_dict = { 11 | 'Binarizer':Binarizer(), 12 | 'MaxAbsScaler':MaxAbsScaler(), 13 | 'MinMaxScaler':MinMaxScaler(), 14 | 'Normalizer':Normalizer(), 15 | 'PolynomialFeatures':PolynomialFeatures(), 16 | 'RobustScaler':RobustScaler(), 17 | 'StandardScaler':StandardScaler(), 18 | 'FastICA':FastICA(), 19 | 'PCA':PCA(), 20 | 'RBFSampler':RBFSampler(), 21 | 'Nystroem':Nystroem(), 22 | 'FeatureAgglomeration':FeatureAgglomeration(), 23 | 'SelectFwe':SelectFwe(), 24 | 'SelectPercentile':SelectPercentile(), 25 | 'VarianceThreshold':VarianceThreshold(), 26 | 'SelectFromModel':SelectFromModel(estimator=ExtraTreesClassifier(n_estimators=100, random_state=324089)), 27 | 'RFE':RFE(estimator=ExtraTreesClassifier(n_estimators=100, random_state=324089)), 28 | } 29 | -------------------------------------------------------------------------------- /read_file.py: -------------------------------------------------------------------------------- 1 | import pandas as pd 2 | import numpy as np 3 | import pdb 4 | 5 | def read_file(filename, label='class', sep=None): 6 | 7 | if filename.split('.')[-1] == 'gz': 8 | compression = 'gzip' 9 | else: 10 | compression = None 11 | 12 | if sep: 13 | input_data = pd.read_csv(filename, sep=sep, compression=compression) 14 | else: 15 | input_data = pd.read_csv(filename, sep=sep, compression=compression, 16 | engine='python') 17 | 18 | input_data.rename(columns={'Label': 'class','Class':'class', 'target':'class'}, 19 | inplace=True) 20 | 21 | feature_names = np.array([x for x in input_data.columns.values if x != label]) 22 | 23 | X = input_data.drop(label, axis=1).values.astype(float) 24 | y = input_data[label].values 25 | 26 | assert(X.shape[1] == feature_names.shape[0]) 27 | 28 | return X, y, feature_names 29 | -------------------------------------------------------------------------------- /submit_jobs.py: -------------------------------------------------------------------------------- 1 | from glob import glob 2 | import os 3 | import sys 4 | import argparse 5 | 6 | if __name__ == '__main__': 7 | parser = argparse.ArgumentParser(description="Submit long jobs.", 8 | add_help=False) 9 | parser.add_argument('DATA_PATH',type=str) 10 | parser.add_argument('-ml',action='store',dest='mls', type=str, 11 | default='LogisticRegression,RandomForestClassifier,GradientBoostingClassifier,' 12 | 'DecisionTreeClassifier,SVC') 13 | parser.add_argument('--long',action='store_true',dest='LONG', default=False) 14 | parser.add_argument('-n_trials',action='store',dest='TRIALS', default=1) 15 | parser.add_argument('-results',action='store',dest='RDIR',default='../results/',type=str, 16 | help='Results directory') 17 | parser.add_argument('--lsf', action='store_true', dest='LSF', default=False, 18 | help='Run on an LSF HPC (using bsub commands)') 19 | parser.add_argument('-m',action='store',dest='M',default=4096,type=int, 20 | help='LSF memory request and limit (MB)') 21 | args = parser.parse_args() 22 | 23 | datapath = args.DATA_PATH 24 | 25 | if args.LONG: 26 | q = 'moore_long' 27 | else: 28 | q = 'moore_normal' 29 | 30 | lpc_options = '--lsf -q {Q} -m {M} -n_jobs 1'.format(Q=q,M=args.M) 31 | 32 | lsf = '--lsf' if args.LSF else '' 33 | 34 | mls = ','.join([ml for ml in args.mls.split(',')]) 35 | for f in glob(datapath + "/*.csv"): 36 | jobline = ('python analyze.py {DATA} ' 37 | '-ml {ML} ' 38 | '-results {RDIR} -n_trials {NT} ' 39 | '-search grid {LPC} {LSF}').format(DATA=f, 40 | LPC=lpc_options, 41 | ML=mls, 42 | RDIR=args.RDIR, 43 | NT=args.TRIALS, 44 | LSF=lsf) 45 | print(jobline) 46 | os.system(jobline) 47 | 48 | -------------------------------------------------------------------------------- /utils.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | from read_file import read_file 3 | from eli5.sklearn import PermutationImportance 4 | import matplotlib.pyplot as plt 5 | from sklearn.metrics import roc_curve, auc 6 | import pdb 7 | 8 | def feature_importance(save_file, model, model_name, feature_names, training_features, training_classes, random_state, 9 | preps, prep_params, clf_name, clf_params): 10 | """ prints feature importance information for a trained estimator (model)""" 11 | coefs = compute_imp_score(model, model_name, training_features, training_classes, random_state) 12 | 13 | if len(coefs) < len(feature_names): # there must be a feature selection strategy 14 | name = [k for k,v in model.named_steps.items() if getattr(v,'get_support',None)][0] 15 | print('name:',name) 16 | fn = np.asarray(feature_names) 17 | sel_feature_names = fn[model.named_steps[name].get_support()] 18 | j = 0 19 | new_coefs=np.zeros(fn.shape) 20 | for i,f in enumerate(feature_names): 21 | if f in sel_feature_names: 22 | new_coefs[i] = coefs[j] 23 | j = j+1 24 | else: 25 | new_coefs[i] = 0 26 | coefs = new_coefs 27 | assert(len(coefs)==len(feature_names)) 28 | # plot_imp_score(save_file, coefs, feature_names, random_state) 29 | 30 | out_text='' 31 | # algorithm seed feature score 32 | for i,c in enumerate(coefs): 33 | out_text += '\t'.join([preps, 34 | prep_params, 35 | clf_name, 36 | clf_params, 37 | str(random_state), 38 | feature_names[i], 39 | str(c)])+'\n' 40 | 41 | with open(save_file.split('.')[0] + '.imp_score','a') as out: 42 | out.write(out_text) 43 | 44 | def compute_imp_score(model, model_name, training_features, training_classes, random_state): 45 | clf = model.named_steps[model_name] 46 | # pdb.set_trace() 47 | if hasattr(clf, 'coef_'): 48 | coefs = np.abs(clf.coef_.flatten()) 49 | 50 | else: 51 | coefs = getattr(clf, 'feature_importances_', None) 52 | if coefs is None: 53 | perm = PermutationImportance( 54 | estimator=model, 55 | n_iter=5, 56 | random_state=random_state, 57 | refit=False 58 | ) 59 | perm.fit(training_features, training_classes) 60 | coefs = perm.feature_importances_ 61 | 62 | 63 | #return (coefs-np.min(coefs))/(np.max(coefs)-np.min(coefs)) 64 | return coefs/np.sum(coefs) 65 | 66 | # def plot_imp_score(save_file, coefs, feature_names, seed): 67 | # # plot bar charts for top 10 importanct features 68 | # num_bar = min(10, len(coefs)) 69 | # indices = np.argsort(coefs)[-num_bar:] 70 | # h=plt.figure() 71 | # plt.title("Feature importances") 72 | # plt.barh(range(num_bar), coefs[indices], color="r", align="center") 73 | # plt.yticks(range(num_bar), feature_names[indices]) 74 | # plt.ylim([-1, num_bar]) 75 | # h.tight_layout() 76 | # plt.savefig(save_file.split('.')[0] + '_imp_score_' + str(seed) + '.pdf') 77 | 78 | ######################################################################################### ROC Curve 79 | 80 | def roc(save_file, model, y_true, probabilities, random_state, preps, prep_params, clf_name, clf_params): 81 | """prints receiver operator chacteristic curve data""" 82 | 83 | # pdb.set_trace() 84 | fpr,tpr,_ = roc_curve(y_true, probabilities) 85 | 86 | AUC = auc(fpr,tpr) 87 | model_name = save_file.split('/')[-1][:-4] 88 | # print results 89 | out_text='' 90 | for f,t in zip(fpr,tpr): 91 | out_text += '\t'.join([preps, 92 | prep_params, 93 | clf_name, 94 | clf_params, 95 | str(random_state), 96 | str(f), 97 | str(t), 98 | str(AUC)])+'\n' 99 | 100 | with open(save_file.split('.')[0] + '.roc','a') as out: 101 | out.write(out_text) 102 | 103 | 104 | --------------------------------------------------------------------------------