├── .gitattributes ├── .gitignore ├── LICENSE.txt ├── README.md ├── data └── processed │ └── .gitkeep ├── models └── .gitkeep ├── notebooks ├── .gitkeep └── 1.0-dsp-LSTMsEnsemle.ipynb ├── setup.py ├── src ├── __init__.py └── data │ ├── .gitkeep │ ├── __init__.py │ └── dataset.py └── test_environment.py /.gitattributes: -------------------------------------------------------------------------------- 1 | * linguist-vendored 2 | *.py linguist-vendored=false 3 | -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | 5 | # C extensions 6 | *.so 7 | 8 | # Distribution / packaging 9 | .Python 10 | env/ 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | *.egg-info/ 23 | .installed.cfg 24 | *.egg 25 | 26 | # PyInstaller 27 | # Usually these files are written by a python script from a template 28 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 29 | *.manifest 30 | *.spec 31 | 32 | # Installer logs 33 | pip-log.txt 34 | pip-delete-this-directory.txt 35 | 36 | # Unit test / coverage reports 37 | htmlcov/ 38 | .tox/ 39 | .coverage 40 | .coverage.* 41 | .cache 42 | nosetests.xml 43 | coverage.xml 44 | *.cover 45 | 46 | # Translations 47 | *.mo 48 | *.pot 49 | 50 | # Django stuff: 51 | *.log 52 | 53 | # Sphinx documentation 54 | docs/_build/ 55 | 56 | # PyBuilder 57 | target/ 58 | 59 | # DotEnv configuration 60 | .env 61 | 62 | # Database 63 | *.db 64 | *.rdb 65 | 66 | # Pycharm 67 | .idea 68 | 69 | # VS Code 70 | .vscode/ 71 | 72 | # Spyder 73 | .spyproject/ 74 | 75 | # Jupyter NB Checkpoints 76 | .ipynb_checkpoints/ 77 | 78 | # exclude data from source control by default 79 | #/data/ 80 | 81 | # exclude models from source control 82 | /models/ 83 | # Mac OS-specific storage files 84 | .DS_Store 85 | 86 | # vim 87 | *.swp 88 | *.swo 89 | 90 | # Mypy cache 91 | .mypy_cache/ 92 | -------------------------------------------------------------------------------- /LICENSE.txt: -------------------------------------------------------------------------------- 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 | ## Overview 2 | 3 | This repository includes the Pytorch implementation of the paper "Ensembles of Deep LSTM Learners for Activity Recognition using Wearables" by Yu Guan and Thomas Plötz, which is available at: https://doi.org/10.1145/3090076 4 | 5 | You can find the authors' original implementation in tensorflow at: https://github.com/tploetz/LSTMEnsemble4HAR 6 | 7 | To run the code, open up "1.0-dsp-LSTMsEnsemble.ipynb" jupyter notebook under notebooks folder and follow the step by step instructions. 8 | 9 | ## Dependencies 10 | 11 | - Python 3
12 | - Pytorch 13 | 14 | Project Organization 15 | ------------ 16 | 17 | ├── LICENSE 18 | ├── README.md <- The top-level README for developers using this project. 19 | ├── data    20 | │   └── processed <- The final, canonical data sets for modeling. 21 | │ 22 | ├── models <- Trained models 23 | │ 24 | ├── notebooks <- Jupyter notebooks. 25 | │ └── 1.0-dsp-LSTMsEnsemble.ipynb <-- Full Pipeline in a step by step manner 26 | │ 27 | └── src <- Source code for use in this project. 28 |    ├── __init__.py <- Makes src a Python module 29 | │ 30 |    └── data <- Scripts to download or generate data 31 |      └── dataset.py 32 | 33 | -------- 34 | 35 |

Project based on the cookiecutter data science project template. #cookiecutterdatascience

36 | -------------------------------------------------------------------------------- /data/processed/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dspanah/Sensor-Based-Human-Activity-Recognition-LSTMsEnsemble-Pytorch/9ea07659f284a153cb874b3f07ce73924c8c0cc2/data/processed/.gitkeep -------------------------------------------------------------------------------- /models/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dspanah/Sensor-Based-Human-Activity-Recognition-LSTMsEnsemble-Pytorch/9ea07659f284a153cb874b3f07ce73924c8c0cc2/models/.gitkeep -------------------------------------------------------------------------------- /notebooks/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dspanah/Sensor-Based-Human-Activity-Recognition-LSTMsEnsemble-Pytorch/9ea07659f284a153cb874b3f07ce73924c8c0cc2/notebooks/.gitkeep -------------------------------------------------------------------------------- /notebooks/1.0-dsp-LSTMsEnsemle.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "cells": [ 3 | { 4 | "cell_type": "markdown", 5 | "metadata": {}, 6 | "source": [ 7 | "### Setup" 8 | ] 9 | }, 10 | { 11 | "cell_type": "code", 12 | "execution_count": 1, 13 | "metadata": {}, 14 | "outputs": [], 15 | "source": [ 16 | "import warnings\n", 17 | "warnings.filterwarnings('ignore')\n", 18 | "\n", 19 | "import torch\n", 20 | "from torch import nn\n", 21 | "import torch.nn.functional as F\n", 22 | "\n", 23 | "import numpy as np\n", 24 | "import pickle\n", 25 | "import pandas as pd\n", 26 | "\n", 27 | "import sklearn.metrics as metrics" 28 | ] 29 | }, 30 | { 31 | "cell_type": "code", 32 | "execution_count": 2, 33 | "metadata": {}, 34 | "outputs": [], 35 | "source": [ 36 | "import os\n", 37 | "import sys\n", 38 | "\n", 39 | "# add the 'src' directory as one where we can import modules\n", 40 | "src_dir = os.path.join(os.getcwd(), os.pardir, 'src')\n", 41 | "sys.path.append(src_dir)\n", 42 | "\n", 43 | "from data.dataset import loadingDB" 44 | ] 45 | }, 46 | { 47 | "cell_type": "code", 48 | "execution_count": 3, 49 | "metadata": {}, 50 | "outputs": [], 51 | "source": [ 52 | "#create results folder\n", 53 | "!mkdir -p ../models/results" 54 | ] 55 | }, 56 | { 57 | "cell_type": "markdown", 58 | "metadata": {}, 59 | "source": [ 60 | "### Download Datasets" 61 | ] 62 | }, 63 | { 64 | "cell_type": "code", 65 | "execution_count": 4, 66 | "metadata": {}, 67 | "outputs": [ 68 | { 69 | "name": "stdout", 70 | "output_type": "stream", 71 | "text": [ 72 | "Requirement already satisfied: gdown in /data/anaconda/envs/py35/lib/python3.5/site-packages (3.8.1)\n", 73 | "Requirement already satisfied: filelock in /data/anaconda/envs/py35/lib/python3.5/site-packages (from gdown) (3.0.4)\n", 74 | "Requirement already satisfied: tqdm in /data/anaconda/envs/py35/lib/python3.5/site-packages (from gdown) (4.28.1)\n", 75 | "Requirement already satisfied: requests in /data/anaconda/envs/py35/lib/python3.5/site-packages (from gdown) (2.18.4)\n", 76 | "Requirement already satisfied: six in /data/anaconda/envs/py35/lib/python3.5/site-packages (from gdown) (1.11.0)\n", 77 | "Requirement already satisfied: chardet<3.1.0,>=3.0.2 in /data/anaconda/envs/py35/lib/python3.5/site-packages (from requests->gdown) (3.0.4)\n", 78 | "Requirement already satisfied: urllib3<1.23,>=1.21.1 in /data/anaconda/envs/py35/lib/python3.5/site-packages (from requests->gdown) (1.22)\n", 79 | "Requirement already satisfied: certifi>=2017.4.17 in /data/anaconda/envs/py35/lib/python3.5/site-packages (from requests->gdown) (2018.8.24)\n", 80 | "Requirement already satisfied: idna<2.7,>=2.5 in /data/anaconda/envs/py35/lib/python3.5/site-packages (from requests->gdown) (2.6)\n", 81 | "\u001b[33mYou are using pip version 18.1, however version 19.0.3 is available.\n", 82 | "You should consider upgrading via the 'pip install --upgrade pip' command.\u001b[0m\n", 83 | "Downloading...\n", 84 | "From: https://drive.google.com/uc?id=1nkAwjp1TRB-wnOYBvlRJS_srv2c6Spz7\n", 85 | "To: /data/home/ml/notebooks/ensemblelstm_pytorch/data/processed/opp.mat\n", 86 | "177MB [00:00, 190MB/s] \n", 87 | "Downloading...\n", 88 | "From: https://drive.google.com/uc?id=1KJ04DWE7nt_PB0Zm9ZaN-Wh-ZYgvBOj-\n", 89 | "To: /data/home/ml/notebooks/ensemblelstm_pytorch/data/processed/pamap2.mat\n", 90 | "140MB [00:00, 150MB/s] \n", 91 | "Downloading...\n", 92 | "From: https://drive.google.com/uc?id=15Q8oV02h2_e94IWJ9rnKLrSCKPCTW5FS\n", 93 | "To: /data/home/ml/notebooks/ensemblelstm_pytorch/data/processed/skoda.mat\n", 94 | "114MB [00:01, 112MB/s] \n" 95 | ] 96 | } 97 | ], 98 | "source": [ 99 | "# run below commands to download datasets from google drive using Gdown tool\n", 100 | "# Alternatively you can manually download datasets from following url and put them in the data folder\n", 101 | "# https://goo.gl/wgEuhu\n", 102 | "\n", 103 | "!pip install gdown\n", 104 | "!mkdir -p ../data/processed\n", 105 | "!gdown https://drive.google.com/uc?id=1nkAwjp1TRB-wnOYBvlRJS_srv2c6Spz7 -O ../data/processed/opp.mat\n", 106 | "!gdown https://drive.google.com/uc?id=1KJ04DWE7nt_PB0Zm9ZaN-Wh-ZYgvBOj- -O ../data/processed/pamap2.mat\n", 107 | "!gdown https://drive.google.com/uc?id=15Q8oV02h2_e94IWJ9rnKLrSCKPCTW5FS -O ../data/processed/skoda.mat" 108 | ] 109 | }, 110 | { 111 | "cell_type": "markdown", 112 | "metadata": {}, 113 | "source": [ 114 | "### Choose Dataset" 115 | ] 116 | }, 117 | { 118 | "cell_type": "code", 119 | "execution_count": 5, 120 | "metadata": {}, 121 | "outputs": [ 122 | { 123 | "name": "stdout", 124 | "output_type": "stream", 125 | "text": [ 126 | "../data/processed/opp.mat\n", 127 | "normalising... zero mean, unit variance\n", 128 | "normalising...X_train, X_valid, X_test... done\n", 129 | "loading the 79-dim matData successfully . . .\n", 130 | "\n", 131 | "Train data shape: inputs(650972, 79), targets (650972,)\n", 132 | "Valid data shape: inputs(32224, 79), targets (32224,)\n", 133 | "Test data shape: inputs(118750, 79), targets (118750,)\n" 134 | ] 135 | } 136 | ], 137 | "source": [ 138 | "#1 is Opportunity , 2 is PAMAP2, 3 is Skoda\n", 139 | "dataset = 1\n", 140 | "\n", 141 | "if dataset == 1:\n", 142 | "\ttrain_x, valid_x, test_x, train_y, valid_y, test_y = loadingDB('../data/processed/', 79)\n", 143 | "\tn_classes = 18\n", 144 | "\tDB = 79\n", 145 | "if dataset == 2:\n", 146 | "\ttrain_x, valid_x, test_x, train_y, valid_y, test_y = loadingDB('../data/processed/', 52)\n", 147 | "\tn_classes = 12\n", 148 | "\tDB = 52\n", 149 | "if dataset == 3:\n", 150 | "\ttrain_x, valid_x, test_x, train_y, valid_y, test_y = loadingDB('../data/processed/', 60)\n", 151 | "\tn_classes = 11\n", 152 | "\tDB = 60\n", 153 | " \n", 154 | "print(\"\\nTrain data shape: inputs{0}, targets {1}\".format(train_x.shape, train_y.shape))\n", 155 | "print(\"Valid data shape: inputs{0}, targets {1}\".format(valid_x.shape, valid_y.shape))\n", 156 | "print(\"Test data shape: inputs{0}, targets {1}\".format(test_x.shape ,test_y.shape))" 157 | ] 158 | }, 159 | { 160 | "cell_type": "markdown", 161 | "metadata": {}, 162 | "source": [ 163 | "### Reshape Validation and Test Data" 164 | ] 165 | }, 166 | { 167 | "cell_type": "code", 168 | "execution_count": 6, 169 | "metadata": {}, 170 | "outputs": [ 171 | { 172 | "name": "stdout", 173 | "output_type": "stream", 174 | "text": [ 175 | "Valid data shape: inputs(1, 32224, 79), targets (1, 32224)\n", 176 | "Test data shape: inputs(1, 118750, 79), targets (1, 118750)\n" 177 | ] 178 | } 179 | ], 180 | "source": [ 181 | "DIM = len(train_x[0])\n", 182 | "TEST_WIN = 5000\n", 183 | "\n", 184 | "valid_bt = 1\n", 185 | "valid_se = len(valid_x)//valid_bt\n", 186 | "valid_x = valid_x[:valid_se*valid_bt,]\n", 187 | "valid_y = np.array(valid_y)\n", 188 | "valid_y = valid_y[:valid_se*valid_bt,]\n", 189 | "valid_x = np.reshape(valid_x, (valid_bt, -1, DB))\n", 190 | "valid_y = np.reshape(valid_y, (valid_bt,-1))\n", 191 | "print(\"Valid data shape: inputs{0}, targets {1}\".format(valid_x.shape, valid_y.shape))\n", 192 | "\n", 193 | "test_bt = 1\n", 194 | "test_se = len(test_x)//test_bt\n", 195 | "test_x = test_x[:test_se*test_bt,]\n", 196 | "test_y = np.array(test_y)\n", 197 | "test_y = test_y[:test_se*test_bt,]\n", 198 | "test_x = np.reshape(test_x, (test_bt, -1, DB))\n", 199 | "test_y = np.reshape(test_y, (test_bt,-1))\n", 200 | "print(\"Test data shape: inputs{0}, targets {1}\".format(test_x.shape ,test_y.shape))" 201 | ] 202 | }, 203 | { 204 | "cell_type": "code", 205 | "execution_count": 7, 206 | "metadata": {}, 207 | "outputs": [], 208 | "source": [ 209 | "def making_training_set(train_x, train_y, batch_size):\n", 210 | " \n", 211 | " seqence_len = len(train_x)//batch_size\n", 212 | " \n", 213 | " # generate random initial position of sampling for each epoch\n", 214 | " indices_start = np.random.randint(low=0, high=len(train_x)-seqence_len, size=(batch_size,))\n", 215 | " \n", 216 | " indices_all_2d = np.zeros((batch_size, seqence_len))\n", 217 | " for i in range(batch_size):\n", 218 | " indices_all_2d[i,:] = np.arange(indices_start[i],indices_start[i]+seqence_len)\n", 219 | " indices_all = np.reshape(indices_all_2d, (-1))\n", 220 | "\n", 221 | " X_train = np.zeros((batch_size, seqence_len, DIM), dtype=np.float32)\n", 222 | " y_train = np.zeros((batch_size, seqence_len), dtype=np.uint8) \n", 223 | " for i in range(batch_size):\n", 224 | " idx_start = indices_start[i]\n", 225 | " idx_end = idx_start+seqence_len\n", 226 | " X_train[i,:,:] = train_x[idx_start:idx_end, :]\n", 227 | " y_train[i,:] = train_y[idx_start:idx_end]\n", 228 | " return X_train, y_train" 229 | ] 230 | }, 231 | { 232 | "cell_type": "markdown", 233 | "metadata": {}, 234 | "source": [ 235 | "### Define the Model" 236 | ] 237 | }, 238 | { 239 | "cell_type": "code", 240 | "execution_count": 8, 241 | "metadata": {}, 242 | "outputs": [], 243 | "source": [ 244 | "class SingleModel(nn.Module):\n", 245 | " \n", 246 | " def __init__(self, n_channels=DB, n_hidden=256, n_layers=2, \n", 247 | " n_classes=n_classes, drop_prob=0.5):\n", 248 | " super(SingleModel, self).__init__()\n", 249 | " \n", 250 | " self.n_layers = n_layers\n", 251 | " self.n_hidden = n_hidden\n", 252 | " self.n_classes = n_classes\n", 253 | " self.drop_prob = drop_prob\n", 254 | " self.n_channels = n_channels\n", 255 | " \n", 256 | " self.lstm = nn.LSTM(n_channels, n_hidden, n_layers, dropout=self.drop_prob)\n", 257 | " self.fc = nn.Linear(n_hidden, n_classes)\n", 258 | " self.dropout = nn.Dropout(drop_prob)\n", 259 | " \n", 260 | " def forward(self, x, hidden, batch_size):\n", 261 | " \n", 262 | " x = x.permute(1, 0, 2)\n", 263 | " x, hidden = self.lstm(x, hidden)\n", 264 | " x = self.dropout(x) \n", 265 | " x = x.contiguous().view(-1, self.n_hidden)\n", 266 | " out = self.fc(x)\n", 267 | " \n", 268 | " return out, hidden\n", 269 | " \n", 270 | " def init_hidden(self, batch_size):\n", 271 | " ''' Initializes hidden state '''\n", 272 | " # Create two new tensors with sizes n_layers x batch_size x n_hidden,\n", 273 | " # initialized to zero, for hidden state and cell state of LSTM\n", 274 | " weight = next(self.parameters()).data\n", 275 | " \n", 276 | " if (train_on_gpu):\n", 277 | " hidden = (weight.new(self.n_layers, batch_size, self.n_hidden).zero_().cuda(),\n", 278 | " weight.new(self.n_layers, batch_size, self.n_hidden).zero_().cuda())\n", 279 | " else:\n", 280 | " hidden = (weight.new(self.n_layers, batch_size, self.n_hidden).zero_(),\n", 281 | " weight.new(self.n_layers, batch_size, self.n_hidden).zero_())\n", 282 | " return hidden\n", 283 | " \n", 284 | "net = SingleModel()" 285 | ] 286 | }, 287 | { 288 | "cell_type": "markdown", 289 | "metadata": {}, 290 | "source": [ 291 | "## Initialize Model Weights" 292 | ] 293 | }, 294 | { 295 | "cell_type": "code", 296 | "execution_count": 9, 297 | "metadata": {}, 298 | "outputs": [ 299 | { 300 | "data": { 301 | "text/plain": [ 302 | "SingleModel(\n", 303 | " (lstm): LSTM(79, 256, num_layers=2, dropout=0.5)\n", 304 | " (fc): Linear(in_features=256, out_features=18, bias=True)\n", 305 | " (dropout): Dropout(p=0.5)\n", 306 | ")" 307 | ] 308 | }, 309 | "execution_count": 9, 310 | "metadata": {}, 311 | "output_type": "execute_result" 312 | } 313 | ], 314 | "source": [ 315 | "def init_weights(m):\n", 316 | " if type(m) == nn.LSTM:\n", 317 | " for name, param in m.named_parameters():\n", 318 | " if 'weight_ih' in name:\n", 319 | " torch.nn.init.orthogonal_(param.data)\n", 320 | " elif 'weight_hh' in name:\n", 321 | " torch.nn.init.orthogonal_(param.data)\n", 322 | " elif 'bias' in name:\n", 323 | " param.data.fill_(0)\n", 324 | " elif type(m) == nn.Linear:\n", 325 | " torch.nn.init.orthogonal_(m.weight)\n", 326 | " m.bias.data.fill_(0)\n", 327 | "net.apply(init_weights) " 328 | ] 329 | }, 330 | { 331 | "cell_type": "code", 332 | "execution_count": 10, 333 | "metadata": {}, 334 | "outputs": [ 335 | { 336 | "name": "stdout", 337 | "output_type": "stream", 338 | "text": [ 339 | "Training on GPU!\n" 340 | ] 341 | } 342 | ], 343 | "source": [ 344 | "# check if GPU is available\n", 345 | "train_on_gpu = torch.cuda.is_available()\n", 346 | "if(train_on_gpu):\n", 347 | " print('Training on GPU!')\n", 348 | "else: \n", 349 | " print('No GPU available, training on CPU; consider making n_epochs very small.')" 350 | ] 351 | }, 352 | { 353 | "cell_type": "markdown", 354 | "metadata": {}, 355 | "source": [ 356 | "## Validate the Model" 357 | ] 358 | }, 359 | { 360 | "cell_type": "code", 361 | "execution_count": 11, 362 | "metadata": {}, 363 | "outputs": [], 364 | "source": [ 365 | "def validation(criterion):\n", 366 | " \n", 367 | " val_accuracy=0\n", 368 | " val_f1score=0\n", 369 | " val_losses = []\n", 370 | " num_val_process = valid_se//TEST_WIN + 1\n", 371 | " val_h = net.init_hidden(valid_bt)\n", 372 | " net.eval()\n", 373 | "\n", 374 | " for j in range(num_val_process):\n", 375 | " start = j*TEST_WIN\n", 376 | " end = np.min((valid_se, start+TEST_WIN))\n", 377 | " \n", 378 | " x = valid_x[:,start:end,:]\n", 379 | " y = valid_y[:,start:end]\n", 380 | "\n", 381 | " inputs, targets = torch.from_numpy(x), torch.from_numpy(y.flatten('F'))\n", 382 | " if(train_on_gpu):\n", 383 | " inputs, targets = inputs.cuda(), targets.cuda()\n", 384 | " \n", 385 | " val_h = tuple([each.data for each in val_h])\n", 386 | " \n", 387 | " output, val_h = net(inputs, val_h, valid_bt)\n", 388 | "\n", 389 | " val_loss = criterion(output, targets.long())\n", 390 | " val_losses.append(val_loss.item())\n", 391 | " \n", 392 | " top_p, top_class = output.topk(1, dim=1)\n", 393 | " equals = top_class == targets.view(*top_class.shape).long()\n", 394 | " val_accuracy += torch.mean(equals.type(torch.FloatTensor))\n", 395 | " val_f1score += metrics.f1_score(top_class.cpu(), targets.view(*top_class.shape).long().cpu(), average='macro')\n", 396 | " \n", 397 | " test_accuracy=0\n", 398 | " test_f1score=0\n", 399 | " test_losses = []\n", 400 | " num_test_process = test_se//TEST_WIN + 1\n", 401 | " test_h = net.init_hidden(test_bt)\n", 402 | " \n", 403 | " for j in range(num_test_process):\n", 404 | " start = j*TEST_WIN\n", 405 | " end = np.min((test_se, start+TEST_WIN))\n", 406 | " \n", 407 | " x = test_x[:,start:end,:]\n", 408 | " y = test_y[:,start:end]\n", 409 | "\n", 410 | " inputs, targets = torch.from_numpy(x), torch.from_numpy(y.flatten('F'))\n", 411 | " if(train_on_gpu):\n", 412 | " inputs, targets = inputs.cuda(), targets.cuda()\n", 413 | " \n", 414 | " test_h = tuple([each.data for each in test_h])\n", 415 | " \n", 416 | " output, test_h = net(inputs, test_h, test_bt)\n", 417 | "\n", 418 | " test_loss = criterion(output, targets.long())\n", 419 | " test_losses.append(test_loss.item())\n", 420 | " \n", 421 | " top_p, top_class = output.topk(1, dim=1)\n", 422 | " equals = top_class == targets.view(*top_class.shape).long()\n", 423 | " test_accuracy += torch.mean(equals.type(torch.FloatTensor))\n", 424 | " test_f1score += metrics.f1_score(top_class.cpu(), targets.view(*top_class.shape).long().cpu(), average='macro')\n", 425 | " \n", 426 | " valid_losses_avg = np.mean(val_losses)\n", 427 | " valid_f1_avg = val_f1score/num_val_process\n", 428 | " print(' '*16 +\"Val Loss: {:.4f}...\".format(valid_losses_avg),\n", 429 | " \"Val Acc: {:.4f}...\".format(val_accuracy/num_val_process),\n", 430 | " \"Val F1: {:.4f}...\".format(valid_f1_avg))\n", 431 | " \n", 432 | " test_losses_avg = np.mean(test_losses)\n", 433 | " test_f1_avg = test_f1score/num_test_process\n", 434 | " print(' '*16 +\"Test Loss: {:.4f}...\".format(test_losses_avg),\n", 435 | " \"Test Acc: {:.4f}...\".format(test_accuracy/num_test_process),\n", 436 | " \"Test F1: {:.4f}...\".format(test_f1_avg))\n", 437 | " \n", 438 | " net.train() # reset to train mode after iterationg through validation data\n", 439 | " \n", 440 | " return valid_losses_avg, test_losses_avg, valid_f1_avg, test_f1_avg" 441 | ] 442 | }, 443 | { 444 | "cell_type": "markdown", 445 | "metadata": {}, 446 | "source": [ 447 | "## Train the Model" 448 | ] 449 | }, 450 | { 451 | "cell_type": "code", 452 | "execution_count": 12, 453 | "metadata": {}, 454 | "outputs": [], 455 | "source": [ 456 | "def train(net, epochs=100, lr=0.001):\n", 457 | " \n", 458 | " opt = torch.optim.Adam(net.parameters(), lr=lr) \n", 459 | " criterion = nn.CrossEntropyLoss()\n", 460 | " \n", 461 | " if(train_on_gpu):\n", 462 | " net.cuda()\n", 463 | " \n", 464 | " train_losses = [] \n", 465 | " results = np.empty([0, 5], dtype=np.float32)\n", 466 | " net.train()\n", 467 | " \n", 468 | " for epoch in range(epochs):\n", 469 | " epoch_loss = 0\n", 470 | " train_loss = 0\n", 471 | " train_sz = 0\n", 472 | " \n", 473 | " #generate random batch size for each epoch\n", 474 | " batch_size = np.random.randint(low=128, high=256, size=1)[0]\n", 475 | " \n", 476 | " # initialize hidden state\n", 477 | " h = net.init_hidden(batch_size) \n", 478 | " \n", 479 | " x_train, y_train = making_training_set(train_x, train_y, batch_size)\n", 480 | " train_len = len(train_x)//batch_size\n", 481 | "\n", 482 | " pos_start = 0\n", 483 | " pos_end = 0\n", 484 | " while pos_end < train_len:\n", 485 | "\n", 486 | " # generate a random window length in each training process\n", 487 | " curr_win_len = np.random.randint(low=16, high=32, size=1)[0]\n", 488 | " \n", 489 | " pos_start = pos_end\n", 490 | " pos_end += curr_win_len\n", 491 | "\n", 492 | " x = x_train[:,pos_start:pos_end,:]\n", 493 | " y = y_train[:,pos_start:pos_end]\n", 494 | " \n", 495 | " inputs, targets = torch.from_numpy(x), torch.from_numpy(y.flatten('F'))\n", 496 | " if(train_on_gpu):\n", 497 | " inputs, targets = inputs.cuda(), targets.cuda()\n", 498 | " \n", 499 | " # Creating new variables for the hidden state, otherwise\n", 500 | " # we'd backprop through the entire training history\n", 501 | " h = tuple([each.data for each in h])\n", 502 | " \n", 503 | " # zero accumulated gradients\n", 504 | " opt.zero_grad() \n", 505 | " \n", 506 | " output, h = net(inputs, h, batch_size)\n", 507 | " \n", 508 | " loss = criterion(output, targets.long())\n", 509 | " \n", 510 | " epoch_loss += loss.item()\n", 511 | " sample_sz = batch_size*curr_win_len\n", 512 | " train_loss += loss.item()*sample_sz\n", 513 | " train_sz += sample_sz\n", 514 | " \n", 515 | " loss.backward()\n", 516 | " opt.step()\n", 517 | " \n", 518 | " #saving the models\n", 519 | " PATH = '../models/'+str(DB)+'_'+str(epoch)+'.pth'\n", 520 | " torch.save(net.state_dict(), PATH)\n", 521 | " \n", 522 | " train_loss_avg = train_loss/train_sz\n", 523 | " print(\"Epoch: {}/{}..\".format(epoch+1, epochs),\n", 524 | " \"Train Loss: {:.4f}\".format(train_loss_avg))\n", 525 | " \n", 526 | " valid_loss, test_loss, valid_f1, test_f1 = validation(criterion)\n", 527 | " \n", 528 | " #saving the results\n", 529 | " epoch_results = np.zeros(5)\n", 530 | " \n", 531 | " epoch_results[0] = train_loss_avg\n", 532 | " epoch_results[1] = valid_loss\n", 533 | " epoch_results[2] = test_loss\n", 534 | " epoch_results[3] = valid_f1\n", 535 | " epoch_results[4] = test_f1\n", 536 | " \n", 537 | " results = np.float32(np.vstack((results, epoch_results)))\n", 538 | " \n", 539 | " PATH = '../models/results/'+str(DB)+'.npy'\n", 540 | " np.save(PATH, results)" 541 | ] 542 | }, 543 | { 544 | "cell_type": "markdown", 545 | "metadata": {}, 546 | "source": [ 547 | "### LSTM Ensemble" 548 | ] 549 | }, 550 | { 551 | "cell_type": "code", 552 | "execution_count": 13, 553 | "metadata": {}, 554 | "outputs": [], 555 | "source": [ 556 | "def lstmEnsemble(n_bestM=20):\n", 557 | "\n", 558 | " PATH = '../models/results/'+str(DB)+'.npy'\n", 559 | " results = np.load(PATH)\n", 560 | "\n", 561 | " valid_col = 3 #third column of results is validation f1 \n", 562 | " idx_set = np.argsort(results[:,valid_col])[::-1] # sort results based on validation f1\n", 563 | "\n", 564 | " best_models = []\n", 565 | " best_models.append(idx_set[:n_bestM]) # store the epoch number of top n models\n", 566 | "\n", 567 | " prob_M = np.zeros((n_bestM, test_y.size, n_classes))\n", 568 | " \n", 569 | " for i in range(n_bestM):\n", 570 | " idx = best_models[0][i]\n", 571 | "\n", 572 | " model = '../models/'+str(DB)+'_'+str(idx)+\".pth\"\n", 573 | " net.load_state_dict(torch.load(model))\n", 574 | " \n", 575 | " if(train_on_gpu):\n", 576 | " net.cuda()\n", 577 | "\n", 578 | " num_test_process = test_se//TEST_WIN + 1\n", 579 | " test_accuracy=0\n", 580 | " test_f1score=0\n", 581 | " test_losses = []\n", 582 | " test_h = net.init_hidden(test_bt)\n", 583 | " prob_2d = np.zeros((test_y.size, n_classes))\n", 584 | "\n", 585 | " net.eval()\n", 586 | " for j in range(num_test_process):\n", 587 | " start = j*TEST_WIN\n", 588 | " end = np.min((test_se, start+TEST_WIN))\n", 589 | "\n", 590 | " x = test_x[:,start:end,:]\n", 591 | " y = test_y[:,start:end]\n", 592 | "\n", 593 | " inputs, targets = torch.from_numpy(x), torch.from_numpy(y.flatten('F'))\n", 594 | " if(train_on_gpu):\n", 595 | " inputs, targets = inputs.cuda(), targets.cuda()\n", 596 | "\n", 597 | " test_h = tuple([each.data for each in test_h])\n", 598 | " output, test_h = net(inputs, test_h, test_bt)\n", 599 | "\n", 600 | " prob_2d[start*test_bt:end*test_bt,:] = F.softmax(output).cpu().detach().numpy()\n", 601 | "\n", 602 | " prob_M[i,:,:] = prob_2d #store predictions of each of the top n models\n", 603 | "\n", 604 | " prob_avg = np.mean(prob_M[:,:,:], axis=0) #model fusion by calculating the average of probabilities \n", 605 | " fused_pred = np.argmax(prob_avg, axis=1)\n", 606 | "\n", 607 | " f1_fused = metrics.f1_score(test_y.flatten(\"F\"), fused_pred, average='macro')\n", 608 | "\n", 609 | " print(\"Ensemble of LSTMs F1-score: {:.4f}\".format(f1_fused))" 610 | ] 611 | }, 612 | { 613 | "cell_type": "code", 614 | "execution_count": 14, 615 | "metadata": {}, 616 | "outputs": [ 617 | { 618 | "name": "stdout", 619 | "output_type": "stream", 620 | "text": [ 621 | "Epoch: 1/100.. Train Loss: 0.8390\n", 622 | " Val Loss: 0.5269... Val Acc: 0.7889... Val F1: 0.2910...\n", 623 | " Test Loss: 0.5022... Test Acc: 0.8366... Test F1: 0.3334...\n", 624 | "Epoch: 2/100.. Train Loss: 0.3975\n", 625 | " Val Loss: 0.4060... Val Acc: 0.8831... Val F1: 0.3982...\n", 626 | " Test Loss: 0.3886... Test Acc: 0.8821... Test F1: 0.4574...\n", 627 | "Epoch: 3/100.. Train Loss: 0.3275\n", 628 | " Val Loss: 0.3904... Val Acc: 0.8763... Val F1: 0.5397...\n", 629 | " Test Loss: 0.4724... Test Acc: 0.8486... Test F1: 0.4925...\n", 630 | "Epoch: 4/100.. Train Loss: 0.2937\n", 631 | " Val Loss: 0.3901... Val Acc: 0.8859... Val F1: 0.4639...\n", 632 | " Test Loss: 0.3738... Test Acc: 0.8935... Test F1: 0.5081...\n", 633 | "Epoch: 5/100.. Train Loss: 0.2654\n", 634 | " Val Loss: 0.3726... Val Acc: 0.8961... Val F1: 0.5175...\n", 635 | " Test Loss: 0.3596... Test Acc: 0.8992... Test F1: 0.5204...\n", 636 | "Epoch: 6/100.. Train Loss: 0.2141\n", 637 | " Val Loss: 0.3341... Val Acc: 0.9085... Val F1: 0.5725...\n", 638 | " Test Loss: 0.3297... Test Acc: 0.8992... Test F1: 0.5766...\n", 639 | "Epoch: 7/100.. Train Loss: 0.1773\n", 640 | " Val Loss: 0.3854... Val Acc: 0.8927... Val F1: 0.5645...\n", 641 | " Test Loss: 0.3293... Test Acc: 0.9085... Test F1: 0.5623...\n", 642 | "Epoch: 8/100.. Train Loss: 0.1997\n", 643 | " Val Loss: 0.3879... Val Acc: 0.8916... Val F1: 0.4668...\n", 644 | " Test Loss: 0.3569... Test Acc: 0.8991... Test F1: 0.5703...\n", 645 | "Epoch: 9/100.. Train Loss: 0.1566\n", 646 | " Val Loss: 0.3497... Val Acc: 0.9099... Val F1: 0.5769...\n", 647 | " Test Loss: 0.2785... Test Acc: 0.9149... Test F1: 0.5980...\n", 648 | "Epoch: 10/100.. Train Loss: 0.1116\n", 649 | " Val Loss: 0.4061... Val Acc: 0.8944... Val F1: 0.5135...\n", 650 | " Test Loss: 0.2971... Test Acc: 0.9194... Test F1: 0.6598...\n", 651 | "Epoch: 11/100.. Train Loss: 0.1178\n", 652 | " Val Loss: 0.4196... Val Acc: 0.8997... Val F1: 0.5870...\n", 653 | " Test Loss: 0.3383... Test Acc: 0.9176... Test F1: 0.6476...\n", 654 | "Epoch: 12/100.. Train Loss: 0.1146\n", 655 | " Val Loss: 0.4196... Val Acc: 0.8948... Val F1: 0.5762...\n", 656 | " Test Loss: 0.3908... Test Acc: 0.9121... Test F1: 0.6465...\n", 657 | "Epoch: 13/100.. Train Loss: 0.1211\n", 658 | " Val Loss: 0.4533... Val Acc: 0.8744... Val F1: 0.4952...\n", 659 | " Test Loss: 0.3871... Test Acc: 0.9063... Test F1: 0.6162...\n", 660 | "Epoch: 14/100.. Train Loss: 0.1021\n", 661 | " Val Loss: 0.3637... Val Acc: 0.9124... Val F1: 0.6287...\n", 662 | " Test Loss: 0.3410... Test Acc: 0.9120... Test F1: 0.6372...\n", 663 | "Epoch: 15/100.. Train Loss: 0.0714\n", 664 | " Val Loss: 0.4234... Val Acc: 0.9056... Val F1: 0.4874...\n", 665 | " Test Loss: 0.3762... Test Acc: 0.9126... Test F1: 0.6396...\n", 666 | "Epoch: 16/100.. Train Loss: 0.0869\n", 667 | " Val Loss: 0.4034... Val Acc: 0.9141... Val F1: 0.6177...\n", 668 | " Test Loss: 0.3404... Test Acc: 0.9200... Test F1: 0.6680...\n", 669 | "Epoch: 17/100.. Train Loss: 0.0601\n", 670 | " Val Loss: 0.4001... Val Acc: 0.9155... Val F1: 0.5960...\n", 671 | " Test Loss: 0.4121... Test Acc: 0.9118... Test F1: 0.6170...\n", 672 | "Epoch: 18/100.. Train Loss: 0.0743\n", 673 | " Val Loss: 0.4589... Val Acc: 0.9080... Val F1: 0.5915...\n", 674 | " Test Loss: 0.4085... Test Acc: 0.9122... Test F1: 0.6323...\n", 675 | "Epoch: 19/100.. Train Loss: 0.0845\n", 676 | " Val Loss: 0.4628... Val Acc: 0.9039... Val F1: 0.5092...\n", 677 | " Test Loss: 0.4682... Test Acc: 0.8632... Test F1: 0.6095...\n", 678 | "Epoch: 20/100.. Train Loss: 0.0690\n", 679 | " Val Loss: 0.5738... Val Acc: 0.8736... Val F1: 0.5007...\n", 680 | " Test Loss: 0.4264... Test Acc: 0.9129... Test F1: 0.6186...\n", 681 | "Epoch: 21/100.. Train Loss: 0.0654\n", 682 | " Val Loss: 0.4631... Val Acc: 0.9069... Val F1: 0.5346...\n", 683 | " Test Loss: 0.4182... Test Acc: 0.9137... Test F1: 0.6277...\n", 684 | "Epoch: 22/100.. Train Loss: 0.0714\n", 685 | " Val Loss: 0.4740... Val Acc: 0.8898... Val F1: 0.4688...\n", 686 | " Test Loss: 0.4659... Test Acc: 0.9065... Test F1: 0.5725...\n", 687 | "Epoch: 23/100.. Train Loss: 0.0682\n", 688 | " Val Loss: 0.4795... Val Acc: 0.9147... Val F1: 0.6209...\n", 689 | " Test Loss: 0.4230... Test Acc: 0.9152... Test F1: 0.6572...\n", 690 | "Epoch: 24/100.. Train Loss: 0.0494\n", 691 | " Val Loss: 0.4987... Val Acc: 0.9062... Val F1: 0.5486...\n", 692 | " Test Loss: 0.4684... Test Acc: 0.8978... Test F1: 0.6448...\n", 693 | "Epoch: 25/100.. Train Loss: 0.0543\n", 694 | " Val Loss: 0.4724... Val Acc: 0.9095... Val F1: 0.5928...\n", 695 | " Test Loss: 0.4263... Test Acc: 0.9111... Test F1: 0.6180...\n", 696 | "Epoch: 26/100.. Train Loss: 0.0517\n", 697 | " Val Loss: 0.4634... Val Acc: 0.9103... Val F1: 0.6031...\n", 698 | " Test Loss: 0.4635... Test Acc: 0.9095... Test F1: 0.6693...\n", 699 | "Epoch: 27/100.. Train Loss: 0.0491\n", 700 | " Val Loss: 0.4650... Val Acc: 0.9135... Val F1: 0.5801...\n", 701 | " Test Loss: 0.4707... Test Acc: 0.9161... Test F1: 0.6549...\n", 702 | "Epoch: 28/100.. Train Loss: 0.0678\n", 703 | " Val Loss: 0.4900... Val Acc: 0.9044... Val F1: 0.5311...\n", 704 | " Test Loss: 0.4726... Test Acc: 0.9089... Test F1: 0.6212...\n", 705 | "Epoch: 29/100.. Train Loss: 0.0458\n", 706 | " Val Loss: 0.5327... Val Acc: 0.8998... Val F1: 0.5169...\n", 707 | " Test Loss: 0.4677... Test Acc: 0.9179... Test F1: 0.6499...\n", 708 | "Epoch: 30/100.. Train Loss: 0.0350\n", 709 | " Val Loss: 0.5465... Val Acc: 0.9048... Val F1: 0.5239...\n", 710 | " Test Loss: 0.4608... Test Acc: 0.9161... Test F1: 0.6590...\n", 711 | "Epoch: 31/100.. Train Loss: 0.0398\n", 712 | " Val Loss: 0.5090... Val Acc: 0.9052... Val F1: 0.6064...\n", 713 | " Test Loss: 0.4757... Test Acc: 0.9139... Test F1: 0.6580...\n", 714 | "Epoch: 32/100.. Train Loss: 0.0394\n", 715 | " Val Loss: 0.5186... Val Acc: 0.9041... Val F1: 0.5306...\n", 716 | " Test Loss: 0.4998... Test Acc: 0.9129... Test F1: 0.6201...\n", 717 | "Epoch: 33/100.. Train Loss: 0.0470\n", 718 | " Val Loss: 0.4870... Val Acc: 0.8990... Val F1: 0.5024...\n", 719 | " Test Loss: 0.4713... Test Acc: 0.9117... Test F1: 0.6100...\n", 720 | "Epoch: 34/100.. Train Loss: 0.0570\n", 721 | " Val Loss: 0.4668... Val Acc: 0.8989... Val F1: 0.4710...\n", 722 | " Test Loss: 0.4454... Test Acc: 0.9122... Test F1: 0.5905...\n", 723 | "Epoch: 35/100.. Train Loss: 0.0463\n", 724 | " Val Loss: 0.5543... Val Acc: 0.9007... Val F1: 0.4955...\n", 725 | " Test Loss: 0.5177... Test Acc: 0.9077... Test F1: 0.6303...\n", 726 | "Epoch: 36/100.. Train Loss: 0.0434\n", 727 | " Val Loss: 0.5069... Val Acc: 0.9031... Val F1: 0.5328...\n", 728 | " Test Loss: 0.4724... Test Acc: 0.9125... Test F1: 0.6313...\n", 729 | "Epoch: 37/100.. Train Loss: 0.0278\n", 730 | " Val Loss: 0.5623... Val Acc: 0.9062... Val F1: 0.5366...\n", 731 | " Test Loss: 0.5464... Test Acc: 0.9087... Test F1: 0.6618...\n", 732 | "Epoch: 38/100.. Train Loss: 0.0262\n", 733 | " Val Loss: 0.5621... Val Acc: 0.9053... Val F1: 0.5308...\n", 734 | " Test Loss: 0.5563... Test Acc: 0.9151... Test F1: 0.6346...\n", 735 | "Epoch: 39/100.. Train Loss: 0.0374\n", 736 | " Val Loss: 0.5276... Val Acc: 0.9084... Val F1: 0.5375...\n", 737 | " Test Loss: 0.5145... Test Acc: 0.9162... Test F1: 0.5897...\n", 738 | "Epoch: 40/100.. Train Loss: 0.0264\n", 739 | " Val Loss: 0.5708... Val Acc: 0.9017... Val F1: 0.5029...\n", 740 | " Test Loss: 0.5479... Test Acc: 0.9139... Test F1: 0.5939...\n", 741 | "Epoch: 41/100.. Train Loss: 0.0358\n", 742 | " Val Loss: 0.6070... Val Acc: 0.8899... Val F1: 0.5066...\n", 743 | " Test Loss: 0.4744... Test Acc: 0.9121... Test F1: 0.6193...\n", 744 | "Epoch: 42/100.. Train Loss: 0.0377\n", 745 | " Val Loss: 0.5602... Val Acc: 0.8958... Val F1: 0.5013...\n", 746 | " Test Loss: 0.5302... Test Acc: 0.9166... Test F1: 0.6272...\n", 747 | "Epoch: 43/100.. Train Loss: 0.0241\n", 748 | " Val Loss: 0.6046... Val Acc: 0.8977... Val F1: 0.5253...\n", 749 | " Test Loss: 0.5265... Test Acc: 0.9156... Test F1: 0.6296...\n", 750 | "Epoch: 44/100.. Train Loss: 0.0319\n", 751 | " Val Loss: 0.5526... Val Acc: 0.9027... Val F1: 0.5382...\n", 752 | " Test Loss: 0.5234... Test Acc: 0.9104... Test F1: 0.6222...\n" 753 | ] 754 | }, 755 | { 756 | "name": "stdout", 757 | "output_type": "stream", 758 | "text": [ 759 | "Epoch: 45/100.. Train Loss: 0.0228\n", 760 | " Val Loss: 0.6316... Val Acc: 0.9009... Val F1: 0.5316...\n", 761 | " Test Loss: 0.5612... Test Acc: 0.9161... Test F1: 0.6715...\n", 762 | "Epoch: 46/100.. Train Loss: 0.0365\n", 763 | " Val Loss: 0.5283... Val Acc: 0.9066... Val F1: 0.5358...\n", 764 | " Test Loss: 0.5178... Test Acc: 0.9082... Test F1: 0.6017...\n", 765 | "Epoch: 47/100.. Train Loss: 0.0297\n", 766 | " Val Loss: 0.5853... Val Acc: 0.9014... Val F1: 0.5289...\n", 767 | " Test Loss: 0.5424... Test Acc: 0.9100... Test F1: 0.6148...\n", 768 | "Epoch: 48/100.. Train Loss: 0.0317\n", 769 | " Val Loss: 0.5766... Val Acc: 0.9005... Val F1: 0.5314...\n", 770 | " Test Loss: 0.4976... Test Acc: 0.9113... Test F1: 0.6497...\n", 771 | "Epoch: 49/100.. Train Loss: 0.0333\n", 772 | " Val Loss: 0.5773... Val Acc: 0.8998... Val F1: 0.5303...\n", 773 | " Test Loss: 0.5750... Test Acc: 0.9102... Test F1: 0.6101...\n", 774 | "Epoch: 50/100.. Train Loss: 0.0261\n", 775 | " Val Loss: 0.5577... Val Acc: 0.9025... Val F1: 0.5327...\n", 776 | " Test Loss: 0.5128... Test Acc: 0.9161... Test F1: 0.6289...\n", 777 | "Epoch: 51/100.. Train Loss: 0.0197\n", 778 | " Val Loss: 0.5386... Val Acc: 0.9079... Val F1: 0.5342...\n", 779 | " Test Loss: 0.5320... Test Acc: 0.9216... Test F1: 0.6352...\n", 780 | "Epoch: 52/100.. Train Loss: 0.0253\n", 781 | " Val Loss: 0.5908... Val Acc: 0.9064... Val F1: 0.5887...\n", 782 | " Test Loss: 0.5309... Test Acc: 0.9165... Test F1: 0.6646...\n", 783 | "Epoch: 53/100.. Train Loss: 0.0260\n", 784 | " Val Loss: 0.5995... Val Acc: 0.9041... Val F1: 0.5187...\n", 785 | " Test Loss: 0.5134... Test Acc: 0.9159... Test F1: 0.6210...\n", 786 | "Epoch: 54/100.. Train Loss: 0.0215\n", 787 | " Val Loss: 0.6257... Val Acc: 0.9032... Val F1: 0.5240...\n", 788 | " Test Loss: 0.5849... Test Acc: 0.9122... Test F1: 0.6228...\n", 789 | "Epoch: 55/100.. Train Loss: 0.0197\n", 790 | " Val Loss: 0.6475... Val Acc: 0.9034... Val F1: 0.5410...\n", 791 | " Test Loss: 0.5873... Test Acc: 0.9156... Test F1: 0.6419...\n", 792 | "Epoch: 56/100.. Train Loss: 0.0214\n", 793 | " Val Loss: 0.6387... Val Acc: 0.9001... Val F1: 0.5820...\n", 794 | " Test Loss: 0.5701... Test Acc: 0.9165... Test F1: 0.6432...\n", 795 | "Epoch: 57/100.. Train Loss: 0.0204\n", 796 | " Val Loss: 0.7012... Val Acc: 0.8977... Val F1: 0.5252...\n", 797 | " Test Loss: 0.6209... Test Acc: 0.9157... Test F1: 0.6542...\n", 798 | "Epoch: 58/100.. Train Loss: 0.0229\n", 799 | " Val Loss: 0.6663... Val Acc: 0.8979... Val F1: 0.5227...\n", 800 | " Test Loss: 0.5812... Test Acc: 0.9165... Test F1: 0.6668...\n", 801 | "Epoch: 59/100.. Train Loss: 0.0217\n", 802 | " Val Loss: 0.6317... Val Acc: 0.8966... Val F1: 0.5164...\n", 803 | " Test Loss: 0.5718... Test Acc: 0.9157... Test F1: 0.6112...\n", 804 | "Epoch: 60/100.. Train Loss: 0.0241\n", 805 | " Val Loss: 0.5715... Val Acc: 0.9026... Val F1: 0.5145...\n", 806 | " Test Loss: 0.5503... Test Acc: 0.9159... Test F1: 0.6466...\n", 807 | "Epoch: 61/100.. Train Loss: 0.0242\n", 808 | " Val Loss: 0.5654... Val Acc: 0.8929... Val F1: 0.4849...\n", 809 | " Test Loss: 0.5528... Test Acc: 0.9069... Test F1: 0.5794...\n", 810 | "Epoch: 62/100.. Train Loss: 0.0247\n", 811 | " Val Loss: 0.5569... Val Acc: 0.9011... Val F1: 0.5408...\n", 812 | " Test Loss: 0.6109... Test Acc: 0.9105... Test F1: 0.6230...\n", 813 | "Epoch: 63/100.. Train Loss: 0.0229\n", 814 | " Val Loss: 0.5658... Val Acc: 0.9017... Val F1: 0.5319...\n", 815 | " Test Loss: 0.6208... Test Acc: 0.9170... Test F1: 0.6676...\n", 816 | "Epoch: 64/100.. Train Loss: 0.0218\n", 817 | " Val Loss: 0.6048... Val Acc: 0.9026... Val F1: 0.6128...\n", 818 | " Test Loss: 0.6206... Test Acc: 0.9142... Test F1: 0.6468...\n", 819 | "Epoch: 65/100.. Train Loss: 0.0317\n", 820 | " Val Loss: 0.5407... Val Acc: 0.9042... Val F1: 0.5489...\n", 821 | " Test Loss: 0.5973... Test Acc: 0.9142... Test F1: 0.6489...\n", 822 | "Epoch: 66/100.. Train Loss: 0.0177\n", 823 | " Val Loss: 0.5379... Val Acc: 0.9043... Val F1: 0.5501...\n", 824 | " Test Loss: 0.6314... Test Acc: 0.9121... Test F1: 0.6118...\n", 825 | "Epoch: 67/100.. Train Loss: 0.0272\n", 826 | " Val Loss: 0.5650... Val Acc: 0.9030... Val F1: 0.5537...\n", 827 | " Test Loss: 0.5932... Test Acc: 0.9141... Test F1: 0.5961...\n", 828 | "Epoch: 68/100.. Train Loss: 0.0220\n", 829 | " Val Loss: 0.6270... Val Acc: 0.8981... Val F1: 0.5182...\n", 830 | " Test Loss: 0.6118... Test Acc: 0.9131... Test F1: 0.6237...\n", 831 | "Epoch: 69/100.. Train Loss: 0.0206\n", 832 | " Val Loss: 0.6016... Val Acc: 0.9034... Val F1: 0.5703...\n", 833 | " Test Loss: 0.6082... Test Acc: 0.9143... Test F1: 0.6046...\n", 834 | "Epoch: 70/100.. Train Loss: 0.0161\n", 835 | " Val Loss: 0.6055... Val Acc: 0.9073... Val F1: 0.5707...\n", 836 | " Test Loss: 0.6109... Test Acc: 0.9200... Test F1: 0.6619...\n", 837 | "Epoch: 71/100.. Train Loss: 0.0229\n", 838 | " Val Loss: 0.5872... Val Acc: 0.8999... Val F1: 0.5473...\n", 839 | " Test Loss: 0.5552... Test Acc: 0.9176... Test F1: 0.6371...\n", 840 | "Epoch: 72/100.. Train Loss: 0.0180\n", 841 | " Val Loss: 0.6116... Val Acc: 0.9046... Val F1: 0.5241...\n", 842 | " Test Loss: 0.6025... Test Acc: 0.9163... Test F1: 0.6142...\n", 843 | "Epoch: 73/100.. Train Loss: 0.0160\n", 844 | " Val Loss: 0.5920... Val Acc: 0.9036... Val F1: 0.5277...\n", 845 | " Test Loss: 0.6283... Test Acc: 0.9126... Test F1: 0.6299...\n", 846 | "Epoch: 74/100.. Train Loss: 0.0355\n", 847 | " Val Loss: 0.5728... Val Acc: 0.8963... Val F1: 0.4892...\n", 848 | " Test Loss: 0.4975... Test Acc: 0.9058... Test F1: 0.5745...\n", 849 | "Epoch: 75/100.. Train Loss: 0.0283\n", 850 | " Val Loss: 0.5797... Val Acc: 0.9062... Val F1: 0.4962...\n", 851 | " Test Loss: 0.5919... Test Acc: 0.9103... Test F1: 0.6321...\n", 852 | "Epoch: 76/100.. Train Loss: 0.0154\n", 853 | " Val Loss: 0.5781... Val Acc: 0.9069... Val F1: 0.5548...\n", 854 | " Test Loss: 0.6355... Test Acc: 0.9086... Test F1: 0.5936...\n", 855 | "Epoch: 77/100.. Train Loss: 0.0164\n", 856 | " Val Loss: 0.6210... Val Acc: 0.9039... Val F1: 0.5484...\n", 857 | " Test Loss: 0.6254... Test Acc: 0.9162... Test F1: 0.6393...\n", 858 | "Epoch: 78/100.. Train Loss: 0.0159\n", 859 | " Val Loss: 0.5762... Val Acc: 0.9080... Val F1: 0.5635...\n", 860 | " Test Loss: 0.6129... Test Acc: 0.9116... Test F1: 0.6081...\n", 861 | "Epoch: 79/100.. Train Loss: 0.0175\n", 862 | " Val Loss: 0.6399... Val Acc: 0.9022... Val F1: 0.5522...\n", 863 | " Test Loss: 0.6394... Test Acc: 0.9159... Test F1: 0.6216...\n", 864 | "Epoch: 80/100.. Train Loss: 0.0239\n", 865 | " Val Loss: 0.5609... Val Acc: 0.9036... Val F1: 0.5464...\n", 866 | " Test Loss: 0.6049... Test Acc: 0.9120... Test F1: 0.6088...\n", 867 | "Epoch: 81/100.. Train Loss: 0.0220\n", 868 | " Val Loss: 0.6115... Val Acc: 0.9039... Val F1: 0.5576...\n", 869 | " Test Loss: 0.6151... Test Acc: 0.9148... Test F1: 0.6293...\n", 870 | "Epoch: 82/100.. Train Loss: 0.0195\n", 871 | " Val Loss: 0.6689... Val Acc: 0.8933... Val F1: 0.5440...\n", 872 | " Test Loss: 0.6707... Test Acc: 0.9145... Test F1: 0.6256...\n", 873 | "Epoch: 83/100.. Train Loss: 0.0203\n", 874 | " Val Loss: 0.5789... Val Acc: 0.9002... Val F1: 0.5479...\n", 875 | " Test Loss: 0.6449... Test Acc: 0.9119... Test F1: 0.6094...\n", 876 | "Epoch: 84/100.. Train Loss: 0.0233\n", 877 | " Val Loss: 0.6400... Val Acc: 0.8958... Val F1: 0.5204...\n", 878 | " Test Loss: 0.5939... Test Acc: 0.9136... Test F1: 0.6155...\n", 879 | "Epoch: 85/100.. Train Loss: 0.0179\n", 880 | " Val Loss: 0.6095... Val Acc: 0.8980... Val F1: 0.5010...\n", 881 | " Test Loss: 0.5634... Test Acc: 0.9153... Test F1: 0.6585...\n", 882 | "Epoch: 86/100.. Train Loss: 0.0198\n", 883 | " Val Loss: 0.6478... Val Acc: 0.8936... Val F1: 0.5330...\n", 884 | " Test Loss: 0.6226... Test Acc: 0.9120... Test F1: 0.6247...\n", 885 | "Epoch: 87/100.. Train Loss: 0.0161\n", 886 | " Val Loss: 0.6346... Val Acc: 0.8980... Val F1: 0.5603...\n", 887 | " Test Loss: 0.6027... Test Acc: 0.9149... Test F1: 0.6144...\n", 888 | "Epoch: 88/100.. Train Loss: 0.0133\n", 889 | " Val Loss: 0.6813... Val Acc: 0.8872... Val F1: 0.5526...\n", 890 | " Test Loss: 0.6362... Test Acc: 0.9122... Test F1: 0.6077...\n" 891 | ] 892 | }, 893 | { 894 | "name": "stdout", 895 | "output_type": "stream", 896 | "text": [ 897 | "Epoch: 89/100.. Train Loss: 0.0179\n", 898 | " Val Loss: 0.6439... Val Acc: 0.8875... Val F1: 0.5213...\n", 899 | " Test Loss: 0.5527... Test Acc: 0.9185... Test F1: 0.6265...\n", 900 | "Epoch: 90/100.. Train Loss: 0.0200\n", 901 | " Val Loss: 0.6183... Val Acc: 0.9041... Val F1: 0.5068...\n", 902 | " Test Loss: 0.5752... Test Acc: 0.9145... Test F1: 0.6301...\n", 903 | "Epoch: 91/100.. Train Loss: 0.0213\n", 904 | " Val Loss: 0.6252... Val Acc: 0.8908... Val F1: 0.4876...\n", 905 | " Test Loss: 0.5530... Test Acc: 0.9134... Test F1: 0.6341...\n", 906 | "Epoch: 92/100.. Train Loss: 0.0167\n", 907 | " Val Loss: 0.6465... Val Acc: 0.8964... Val F1: 0.4959...\n", 908 | " Test Loss: 0.5871... Test Acc: 0.9182... Test F1: 0.6637...\n", 909 | "Epoch: 93/100.. Train Loss: 0.0193\n", 910 | " Val Loss: 0.6026... Val Acc: 0.9029... Val F1: 0.5246...\n", 911 | " Test Loss: 0.5631... Test Acc: 0.9176... Test F1: 0.6132...\n", 912 | "Epoch: 94/100.. Train Loss: 0.0284\n", 913 | " Val Loss: 0.5847... Val Acc: 0.9079... Val F1: 0.5283...\n", 914 | " Test Loss: 0.5564... Test Acc: 0.9158... Test F1: 0.6178...\n", 915 | "Epoch: 95/100.. Train Loss: 0.0178\n", 916 | " Val Loss: 0.6040... Val Acc: 0.9083... Val F1: 0.5784...\n", 917 | " Test Loss: 0.5792... Test Acc: 0.9188... Test F1: 0.6318...\n", 918 | "Epoch: 96/100.. Train Loss: 0.0132\n", 919 | " Val Loss: 0.6475... Val Acc: 0.9076... Val F1: 0.5215...\n", 920 | " Test Loss: 0.5989... Test Acc: 0.9192... Test F1: 0.6234...\n", 921 | "Epoch: 97/100.. Train Loss: 0.0108\n", 922 | " Val Loss: 0.6675... Val Acc: 0.9060... Val F1: 0.5164...\n", 923 | " Test Loss: 0.6051... Test Acc: 0.9203... Test F1: 0.6223...\n", 924 | "Epoch: 98/100.. Train Loss: 0.0115\n", 925 | " Val Loss: 0.6666... Val Acc: 0.9038... Val F1: 0.5458...\n", 926 | " Test Loss: 0.5843... Test Acc: 0.9206... Test F1: 0.6365...\n", 927 | "Epoch: 99/100.. Train Loss: 0.0130\n", 928 | " Val Loss: 0.6844... Val Acc: 0.9005... Val F1: 0.5078...\n", 929 | " Test Loss: 0.5875... Test Acc: 0.9197... Test F1: 0.6262...\n", 930 | "Epoch: 100/100.. Train Loss: 0.0106\n", 931 | " Val Loss: 0.6761... Val Acc: 0.8983... Val F1: 0.4985...\n", 932 | " Test Loss: 0.6110... Test Acc: 0.9181... Test F1: 0.6063...\n" 933 | ] 934 | } 935 | ], 936 | "source": [ 937 | "train(net) # train and save results & models" 938 | ] 939 | }, 940 | { 941 | "cell_type": "code", 942 | "execution_count": 15, 943 | "metadata": {}, 944 | "outputs": [ 945 | { 946 | "name": "stdout", 947 | "output_type": "stream", 948 | "text": [ 949 | "Ensemble of LSTMs F1-score: 0.7230\n" 950 | ] 951 | } 952 | ], 953 | "source": [ 954 | "lstmEnsemble(n_bestM=20)" 955 | ] 956 | } 957 | ], 958 | "metadata": { 959 | "kernelspec": { 960 | "display_name": "Python [conda env:py35]", 961 | "language": "python", 962 | "name": "conda-env-py35-py" 963 | }, 964 | "language_info": { 965 | "codemirror_mode": { 966 | "name": "ipython", 967 | "version": 3 968 | }, 969 | "file_extension": ".py", 970 | "mimetype": "text/x-python", 971 | "name": "python", 972 | "nbconvert_exporter": "python", 973 | "pygments_lexer": "ipython3", 974 | "version": "3.5.5" 975 | } 976 | }, 977 | "nbformat": 4, 978 | "nbformat_minor": 2 979 | } 980 | -------------------------------------------------------------------------------- /setup.py: -------------------------------------------------------------------------------- 1 | from setuptools import find_packages, setup 2 | 3 | setup( 4 | name='src', 5 | packages=find_packages(), 6 | version='0.1.0', 7 | description='Ensemble of LSTM models for Human Activity Recognition', 8 | author='Davoud Shariat Panah', 9 | license='', 10 | ) 11 | -------------------------------------------------------------------------------- /src/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dspanah/Sensor-Based-Human-Activity-Recognition-LSTMsEnsemble-Pytorch/9ea07659f284a153cb874b3f07ce73924c8c0cc2/src/__init__.py -------------------------------------------------------------------------------- /src/data/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dspanah/Sensor-Based-Human-Activity-Recognition-LSTMsEnsemble-Pytorch/9ea07659f284a153cb874b3f07ce73924c8c0cc2/src/data/.gitkeep -------------------------------------------------------------------------------- /src/data/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dspanah/Sensor-Based-Human-Activity-Recognition-LSTMsEnsemble-Pytorch/9ea07659f284a153cb874b3f07ce73924c8c0cc2/src/data/__init__.py -------------------------------------------------------------------------------- /src/data/dataset.py: -------------------------------------------------------------------------------- 1 | # motivated by ensemble of deep lstm learners 2 | import numpy as np 3 | import scipy.io 4 | import pandas as pd 5 | 6 | def loadingDB(fileDir, DB=79): 7 | 8 | if DB==79: 9 | matfile = fileDir+'opp.mat' 10 | print(matfile) 11 | data = scipy.io.loadmat(matfile) 12 | 13 | X_train = np.transpose(data['trainingData']) 14 | X_valid = np.transpose(data['valData']) 15 | X_test = np.transpose(data['testingData']) 16 | print('normalising... zero mean, unit variance') 17 | mn_trn = np.mean(X_train, axis=0) 18 | std_trn = np.std(X_train, axis=0) 19 | X_train = (X_train - mn_trn)/std_trn 20 | X_valid = (X_valid - mn_trn)/std_trn 21 | X_test = (X_test - mn_trn)/std_trn 22 | print('normalising...X_train, X_valid, X_test... done') 23 | y_train = data['trainingLabels'].reshape(-1)-1 24 | y_valid = data['valLabels'].reshape(-1)-1 25 | y_test = data['testingLabels'].reshape(-1)-1 26 | print('loading the 79-dim matData successfully . . .') 27 | 28 | if DB==60: 29 | matfile = fileDir+'skoda.mat' 30 | data = scipy.io.loadmat(matfile) 31 | 32 | X_train = data['X_train'] 33 | X_valid = data['X_valid'] 34 | X_test = data['X_test'] 35 | y_train = data['y_train'].reshape(-1) 36 | y_valid = data['y_valid'].reshape(-1) 37 | y_test = data['y_test'].reshape(-1) 38 | print('the Skoda dataset was normalized to zero-mean, unit variance') 39 | print('loading the 33HZ 60d matData successfully . . .') 40 | 41 | if DB==9: 42 | matfile = fileDir+'FOG.mat' 43 | data = scipy.io.loadmat(matfile) 44 | 45 | X_train = data['X_train'] 46 | X_valid = data['X_valid'] 47 | X_test = data['X_test'] 48 | y_train = data['y_train'].reshape(-1) 49 | y_valid = data['y_valid'].reshape(-1) 50 | y_test = data['y_test'].reshape(-1) 51 | print('binary classification problem . . . ') 52 | print('the FOG dataset was normalized to zero-mean, unit variance') 53 | print('loading the 32HZ FOG 9d matData successfully . . .') 54 | 55 | if DB==52: 56 | matfile = fileDir+'pamap2.mat' 57 | data = scipy.io.loadmat(matfile) 58 | 59 | X_train = data['X_train'] 60 | X_valid = data['X_valid'] 61 | X_test = data['X_test'] 62 | y_train = data['y_train'].reshape(-1) 63 | y_valid = data['y_valid'].reshape(-1) 64 | y_test = data['y_test'].reshape(-1) 65 | print('the PAMAP2 dataset was normalized to zero-mean, unit variance') 66 | print('loading the 33HZ PAMAP2 52d matData successfully . . .') 67 | 68 | X_train = X_train.astype(np.float32) 69 | X_valid = X_valid.astype(np.float32) 70 | X_test = X_test.astype(np.float32) 71 | 72 | y_train = y_train.astype(np.uint8) 73 | y_valid = y_valid.astype(np.uint8) 74 | y_test = y_test.astype(np.uint8) 75 | 76 | return X_train, X_valid, X_test, y_train, y_valid, y_test 77 | -------------------------------------------------------------------------------- /test_environment.py: -------------------------------------------------------------------------------- 1 | import sys 2 | 3 | REQUIRED_PYTHON = "python3" 4 | 5 | 6 | def main(): 7 | system_major = sys.version_info.major 8 | if REQUIRED_PYTHON == "python": 9 | required_major = 2 10 | elif REQUIRED_PYTHON == "python3": 11 | required_major = 3 12 | else: 13 | raise ValueError("Unrecognized python interpreter: {}".format( 14 | REQUIRED_PYTHON)) 15 | 16 | if system_major != required_major: 17 | raise TypeError( 18 | "This project requires Python {}. Found: Python {}".format( 19 | required_major, sys.version)) 20 | else: 21 | print(">>> Development environment passes all tests!") 22 | 23 | 24 | if __name__ == '__main__': 25 | main() 26 | --------------------------------------------------------------------------------