├── .gitignore ├── LICENSE ├── README.md ├── code └── helper_functions.py ├── figs ├── figure_prc.pdf ├── figure_prc_condensed.pdf ├── figure_prc_no_pretraining.pdf ├── testing_distributions.pdf ├── training_distributions.pdf ├── training_distributions_1760.pdf ├── training_distributions_1760_fit.pdf ├── training_distributions_208.pdf ├── training_distributions_208_fit.pdf └── training_distributions_fit_combined.pdf └── notebook └── time-base-detector.ipynb /.gitignore: -------------------------------------------------------------------------------- 1 | # ignore all data files with .pkl or .log extension 2 | *.pkl 3 | *.log 4 | 5 | #ignore .html files 6 | *.html 7 | 8 | # ignore exploratory scripts that no one should EVER SEE 9 | /exploratory 10 | data*/ 11 | *data/ 12 | 13 | # virtual environment 14 | env/ 15 | 16 | # ignore journal 17 | *-journal.md 18 | 19 | # ignore ipynb checkpoints 20 | *.ipynb_checkpoints 21 | 22 | # ignore compressed files 23 | *.zip 24 | 25 | # ignore any .pyc files because they are annoying 26 | *.pyc 27 | 28 | ## ignore hidden os files: (not sure what to put in for windows users ) 29 | *.DS_Store 30 | .DS_Store 31 | 32 | ## ignore latex auxiliary files 33 | *.bbl 34 | *.blg 35 | *.dvi 36 | *.out 37 | *.gz 38 | *.aux 39 | *.out 40 | *.log.txt 41 | *.bak 42 | *.csv 43 | -------------------------------------------------------------------------------- /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 | # can-time-based-ids-benchmark 2 | 3 | This repository contains the source code and Jupyter notebook to replicate the results of the autosec21 paper. 4 | 5 | Before running this code, please create a subfolder called `data` in this repo and dump the content of the [ROAD](https://0xsam.com/road/) dataset in there. 6 | 7 | If you use this code, please cite the following papers: 8 | 9 | @unpublished{Verma:2020:ROAD, 10 | title={{Addressing the Lack of Comparability $\&$ Testing in CAN Intrusion Detection Research: A Comprehensive Guide to CAN IDS Data $\&$ Introduction of the ROAD Dataset}}, 11 | author={Miki E. Verma and Michael D. Iannacone and Robert A. Bridges and Samuel C. Hollifield and Pablo Moriano and Bill Kay and Frank L. Combs}, 12 | year={2022}, 13 | note={arXiv preprint \href{https://arxiv.org/abs/2012.14600}{arXiv:2012.14600}, January 2022}, 14 | eprint={2012.14600}, 15 | archivePrefix={arXiv}, 16 | primaryClass={cs.CR} 17 | } 18 | 19 | @inproceedings{Blevins:2021:Time-Based:CAN:IDS:Benchmark, 20 | title={Time-Based CAN Intrusion Detection Benchmark}, 21 | author={Blevins, Deborah H and Moriano, Pablo and Bridges, Robert A and Verma, Miki E and Iannacone, Michael D and Hollifield, Samuel C}, 22 | booktitle={Proceedings of the Workshop on Automotive and Autonomous Vehicle Security (AutoSec)}, 23 | pages={1--6}, 24 | year={2021} 25 | } 26 | 27 | -------------------------------------------------------------------------------- /code/helper_functions.py: -------------------------------------------------------------------------------- 1 | import json 2 | import pickle 3 | import codecs 4 | import copy 5 | import datetime 6 | import os 7 | 8 | import numpy as np 9 | import os 10 | 11 | import matplotlib.pyplot as plt 12 | #%matplotlib inline 13 | 14 | import pandas as pd 15 | 16 | from sklearn.metrics import confusion_matrix, precision_recall_fscore_support 17 | import scipy.stats 18 | #from generalFunctions import * 19 | import cProfile 20 | from collections import Counter 21 | import bisect 22 | 23 | import re 24 | from tqdm import tqdm 25 | 26 | from sklearn.covariance import EllipticEnvelope 27 | 28 | 29 | # functions for saving/opening objects 30 | def jsonify(obj, out_file): 31 | """ 32 | Inputs: 33 | - obj: the object to be jsonified 34 | - out_file: the file path where obj will be saved 35 | This function saves obj to the path out_file as a json file. 36 | """ 37 | json.dump(obj, codecs.open(out_file, 'w', encoding='utf-8'), separators=(',', ':'), sort_keys=True, indent=4) 38 | 39 | 40 | def unjsonify(in_file): 41 | """ 42 | Input: 43 | -in_file: the file path where the object you want to read in is stored 44 | Output: 45 | -obj: the object you want to read in 46 | """ 47 | obj_text = codecs.open(in_file, 'r', encoding='utf-8').read() 48 | obj = json.loads(obj_text) 49 | return obj 50 | 51 | def picklify(obj, filepath): 52 | """ 53 | Inputs: 54 | - obj: the object to be pickled 55 | - filepath: the file path where obj will be saved 56 | This function pickles obj to the path filepath. 57 | """ 58 | pickle_file = open(filepath, "wb") 59 | pickle.dump(obj, pickle_file) 60 | pickle_file.close() 61 | #print "picklify done" 62 | 63 | 64 | def unpickle(filepath): 65 | """ 66 | Input: 67 | -filepath: the file path where the pickled object you want to read in is stored 68 | Output: 69 | -obj: the object you want to read in 70 | """ 71 | pickle_file = open(filepath, 'rb') 72 | obj = pickle.load(pickle_file) 73 | pickle_file.close() 74 | return obj 75 | 76 | def curtime_str(): 77 | """A string representation of the current time.""" 78 | dt = datetime.datetime.now().time() 79 | return dt.strftime("%H:%M:%S") 80 | 81 | 82 | def update_json_dict(key, value, out_file, overwrite = True): 83 | if not os.path.isfile(out_file): 84 | d = {} 85 | else: 86 | d = unjsonify(out_file) 87 | if key in d and not overwrite: 88 | print("fkey {key} already in {out_file}, skipping...") 89 | return 90 | d[key] = value 91 | jsonify(d, out_file) 92 | 93 | #jsonify(sorted(d.items(), key = lambda x: x[0]), out_file) 94 | 95 | 96 | def make_can_df(log_filepath): 97 | """ 98 | Puts candump data into a dataframe with columns 'time', 'aid', and 'data' 99 | """ 100 | can_df = pd.read_fwf( 101 | log_filepath, delimiter = ' '+ '#' + '('+')', 102 | skiprows = 1,skipfooter=1, 103 | usecols = [0,2,3], 104 | dtype = {0:'float64', 1:str, 2: str}, 105 | names = ['time','aid', 'data'] ) 106 | 107 | #print(can_df) 108 | 109 | can_df.aid = can_df.aid.apply(lambda x: int(x,16)) 110 | can_df.data = can_df.data.apply(lambda x: x.zfill(16)) #pad with 0s on the left for data with dlc < 8 111 | can_df.time = can_df.time - can_df.time.min() 112 | 113 | 114 | #print(can_df) 115 | return can_df[can_df.aid<=0x700] 116 | 117 | def extract_k_bits(num, p=8, k=18): 118 | """ 119 | Extract decimal representation of a hex string 120 | @param num: Hex to be analyzed (str) 121 | @param p: Start of the slice (int). It is 9 for J1939 122 | @param k: Length of the slide (int). It is 18 for J1939 123 | @output : Decimal representation of the slice (int) 124 | """ 125 | 126 | # Convert hex str into a 29 bit representation (J1939 AID) 127 | binary = format(int(num, 16), "029b") 128 | #print(binary) 129 | 130 | # Compute slice indices 131 | end = len(binary) - p 132 | start = end - k + 1 133 | 134 | # Slice the binary string 135 | k_bit_string = binary[start:end+1] 136 | 137 | # Convert binary string to decimal 138 | value = int(k_bit_string, 2) 139 | #print(value) 140 | 141 | return(value) 142 | 143 | def make_1939_df(log_filepath): 144 | """ 145 | Puts candump data into a dataframe with columns 'time', 'PGN', and 'data' 146 | @param log_filepath: Path to the file (str) 147 | @output can_df: Data frame representation of the log file (data frame) 148 | """ 149 | can_df = pd.read_fwf( 150 | log_filepath, delimiter=" " + "#" + "(" + ")", 151 | skiprows=0, skipfooter=0, 152 | usecols=[0, 2, 3], 153 | dtype={0:"float64", 1:str, 2:str}, 154 | names=["time", "aid", "data"] ) 155 | 156 | # Discard rows with any NaN 157 | can_df = can_df[can_df["aid"].notna()] 158 | 159 | # print(can_df.dtypes) 160 | # display(can_df) 161 | 162 | # can_df.aid = can_df.aid.apply(lambda x: int(x,16)) 163 | 164 | can_df.aid = can_df.aid.apply(extract_k_bits) 165 | can_df.data = can_df.data.apply(lambda x: x.zfill(16)) #pad with 0s on the left for data with dlc < 8 166 | can_df.time = can_df.time - can_df.time.min() 167 | 168 | can_df.columns = ["time", "PGN", "data"] 169 | 170 | return can_df 171 | 172 | def extract_numbers(string_to_parse): 173 | """ 174 | Extract the first string number (either int of float) from a string 175 | @param string_to_parse: Input string (str) 176 | @output res: number representation (float) 177 | """ 178 | 179 | temp = re.search(r"\d+\.*\d+", string_to_parse) 180 | 181 | if temp: 182 | res = float(temp.group()) 183 | # Transform to seconds 184 | if res > 1: 185 | res = res/1000 186 | else: 187 | res = "" 188 | 189 | return res 190 | 191 | 192 | def add_time_diff_per_aid_col(df, order_by_time = False): 193 | """ 194 | Sorts df by aid and time and takes time diff between each successive col and puts in col "time_diffs" 195 | Then removes first instance of each aids message 196 | Returns sorted df with new column 197 | """ 198 | 199 | df.sort_values(['aid','time'], inplace=True) 200 | df['time_diffs'] = df['time'].diff() 201 | mask = df.aid == df.aid.shift(1) #get bool mask of to filter out first msg of each group 202 | df = df[mask] 203 | if order_by_time: 204 | df = df.sort_values('time').reset_index() 205 | return df 206 | 207 | 208 | def add_time_diff_per_PGN_col(df, order_by_time = False): 209 | """ 210 | Sorts df by aid and time and takes time diff between each successive col and puts in col "time_diffs" 211 | Then removes first instance of each aids message 212 | Returns sorted df with new column 213 | """ 214 | 215 | df.sort_values(['PGN','time'], inplace=True) 216 | df['time_diffs'] = df['time'].diff() 217 | mask = df.PGN == df.PGN.shift(1) #get bool mask of to filter out first msg of each group 218 | df = df[mask] 219 | if order_by_time: 220 | df = df.sort_values('time').reset_index() 221 | return df 222 | 223 | 224 | def get_injection_interval(df, injection_aid, injection_data_str, max_injection_t_delta=1): 225 | """ 226 | Compute time intervals where attacks were injected based on aid and payload 227 | @param df: testing df to be analyzed (dataframe) 228 | @param injection_aid: aid that injects the attack (int) 229 | @param injection_data_str: payload of the attack (str) 230 | @param max_injection_t_delta: minimum separation between attacks (int) 231 | @output injection_intervals: list of intervals where the attacks were injected (list) 232 | """ 233 | 234 | # Construct a regular expression to identify the payload 235 | injection_data_str = injection_data_str.replace("X", ".") 236 | 237 | attack_messages_df = df[(df.aid==injection_aid) & (df.data.str.contains(injection_data_str))] # get subset of attack messages 238 | #print(attack_messages_df) 239 | 240 | if len(attack_messages_df) == 0: 241 | print("message not found") 242 | return None 243 | 244 | # Assuming that attacks are injected with a diferrence more than i seconds 245 | inj_period_times = np.split(np.array(attack_messages_df.time), 246 | np.where(attack_messages_df.time.diff()>max_injection_t_delta)[0]) 247 | 248 | # Pack the intervals 249 | injection_intervals = [(time_arr[0], time_arr[-1]) 250 | for time_arr in inj_period_times if len(time_arr)>1] 251 | 252 | return injection_intervals 253 | 254 | 255 | def add_actual_attack_col(df, intervals, aid, payload): 256 | """ 257 | Adds column to df to indicate which signals were part of attack 258 | """ 259 | 260 | if aid != "XXX": 261 | df['actual_attack'] = df.time.apply(lambda x: sum(x >= intvl[0] and x <= intvl[1] for intvl in intervals ) >= 1) & (df.aid == aid) 262 | 263 | else: 264 | df['actual_attack'] = df.time.apply(lambda x: sum(x >= intvl[0] and x <= intvl[1] for intvl in intervals ) >= 1) & (df.data == payload) 265 | return df 266 | 267 | 268 | def add_kde_val_col(df, d): 269 | """ 270 | Adds column to df with the value of the kde at each time_diff in the df 271 | """ 272 | # df['kde_val'] = df.apply(lambda row: d[row.aid]['kde'].evaluate(row.time_diffs)[0], axis=1) 273 | new_column = np.concatenate([d[aid]["kde"].evaluate(df[df.aid == aid].time_diffs.values) for aid in tqdm(df.aid.unique())]) 274 | df['kde_val'] = new_column 275 | 276 | return df 277 | 278 | 279 | def add_gauss_val_col(df, d): 280 | """ 281 | Adds column to df with the value of the Guassian approximation at each time_diff in the df 282 | """ 283 | # df['gauss_val'] = df.apply(lambda row: d[row.aid]['gauss'].pdf(row.time_diffs), axis = 1) 284 | new_column = np.concatenate([d[aid]["gauss"].pdf(df[df.aid == aid].time_diffs.values) for aid in tqdm(df.aid.unique())]) 285 | df['gauss_val'] = new_column 286 | return df 287 | 288 | 289 | def train(df, aid): 290 | """ 291 | Returns a dictionary the aid including the mean of its time_diffs, standard deviation of its time_diffs 292 | and KDE of its time_diffs 293 | """ 294 | time_diffs = df[df.aid==aid].time_diffs.values 295 | print("before: ", len(time_diffs)) 296 | 297 | # identify outliers in the dataset 298 | ee = EllipticEnvelope(contamination=0.0001, support_fraction=0.999) # support_fraction=0.99 299 | inliers = ee.fit_predict(time_diffs.reshape(-1, 1)) 300 | 301 | # select all rows that are not outliers 302 | mask = inliers != -1 303 | outliers = sum(mask == False) 304 | print("outliers: ", outliers, 100*outliers/len(time_diffs)) 305 | 306 | time_diffs = time_diffs[mask] 307 | # summarize the shape of the updated training dataset 308 | print("after: ", len(time_diffs)) 309 | 310 | aid_dict = {'mu': time_diffs.mean(), 'std': time_diffs.std(), 'kde': scipy.stats.gaussian_kde(time_diffs), 'gauss': scipy.stats.norm(loc = time_diffs.mean(), scale = time_diffs.std())} 311 | return aid_dict 312 | 313 | def train_no_preprocessing(df, aid): 314 | """ 315 | Returns a dictionary the aid including the mean of its time_diffs, standard deviation of its time_diffs 316 | and KDE of its time_diffs 317 | """ 318 | time_diffs = df[df.aid==aid].time_diffs.values 319 | # print("before: ", len(time_diffs)) 320 | 321 | # # identify outliers in the dataset 322 | # ee = EllipticEnvelope(contamination=0.0001, support_fraction=0.999) # support_fraction=0.99 323 | # inliers = ee.fit_predict(time_diffs.reshape(-1, 1)) 324 | 325 | # # select all rows that are not outliers 326 | # mask = inliers != -1 327 | # outliers = sum(mask == False) 328 | # print("outliers: ", outliers, 100*outliers/len(time_diffs)) 329 | 330 | # time_diffs = time_diffs[mask] 331 | # # summarize the shape of the updated training dataset 332 | # print("after: ", len(time_diffs)) 333 | 334 | aid_dict = {'mu': time_diffs.mean(), 'std': time_diffs.std(), 'kde': scipy.stats.gaussian_kde(time_diffs), 'gauss': scipy.stats.norm(loc = time_diffs.mean(), scale = time_diffs.std())} 335 | return aid_dict 336 | 337 | 338 | def train_J1939(df, PGN): 339 | """ 340 | Returns a dictionary the aid including the mean of its time_diffs, standard deviation of its time_diffs 341 | and KDE of its time_diffs 342 | """ 343 | time_diffs = df[df["PGN"] == PGN].time_diffs 344 | aid_dict = {'mu': time_diffs.mean(), 'std': time_diffs.std()} 345 | return aid_dict 346 | 347 | 348 | def y_threshold_kde(dict, aid, p): 349 | """ 350 | Determines the approximate y value at which the KDE of the aid has the desired p value 351 | """ 352 | pvs = [] 353 | mu = dict[aid]['mu'] 354 | std = dict[aid]['std'] 355 | x = np.arange(mu - 5*std, mu + 5*std, 10*std/1000) 356 | #y = [dict[aid]['kde'].evaluate(i_x) for i_x in x] 357 | y = dict[aid]['kde'].evaluate(x) 358 | for i, i_x in enumerate(x): 359 | pvs.append(sum([j for j in y if j <= y[i] ]) * 10*std/1000) 360 | if np.where(np.array(pvs) <= p)[0] != []: 361 | y_threshold = np.max(np.array(y)[np.where(np.array(pvs) <= p)[0]]) 362 | else: y_threshold = 0 363 | dict[aid]['y_thresholds_kde'][p] = y_threshold 364 | 365 | 366 | def y_threshold_gauss(dict, aid, p): 367 | """ 368 | Determines the approximate y value at which the Gaussian approximation of the aid has the desired p value 369 | """ 370 | 371 | pvs = [] 372 | mu = dict[aid]['mu'] 373 | std = dict[aid]['std'] 374 | x = np.arange(mu - 5*std, mu + 5*std, 10*std/1000) 375 | # y = [dict[aid]['gauss'].pdf(i_x) for i_x in x] 376 | y = dict[aid]['gauss'].pdf(x) 377 | for i, i_x in enumerate(x): 378 | pvs.append(sum([j for j in y if j <= y[i] ]) * 10*std/1000) 379 | y_threshold = np.max(np.array(y)[np.where(np.array(pvs) <= p)[0]]) 380 | dict[aid]['y_thresholds_gauss'][p] = y_threshold 381 | 382 | 383 | def alert_by_mean(df, d): 384 | """ 385 | Adds column to df to indicate when time_diff is less than half the mean time_diff for the aid 386 | """ 387 | df['predicted_attack'] = df.apply(lambda row: row.time_diffs <= (d[row.aid]['mu']/2), axis = 1) 388 | return df 389 | 390 | 391 | def alert_by_mean_various_p(df, d, p): 392 | """ 393 | Adds column to df to indicate when time_diff is less than half the mean time_diff for the aid 394 | """ 395 | df['predicted_attack'] = df.apply(lambda row: row.time_diffs <= (p*d[row.aid]['mu']), axis=1) 396 | return df 397 | 398 | 399 | def alert_by_mean_2(df, dict,n): 400 | """ 401 | Same as alert_by_mean but slower 402 | """ 403 | predicted_attack = [] 404 | for i in range(df.count()[0]): 405 | aid = df.iloc[i].aid 406 | time_diff = df.iloc[i].time_diffs 407 | if aid in dict.keys(): 408 | if time_diff <= (dict[aid]['mu'])/n: 409 | predicted_attack.append(True) 410 | else: 411 | predicted_attack.append(False) 412 | else: 413 | #print("aid %s not seen in training; no attacks predicted" %(aid)) 414 | predicted_attack.append(False) 415 | df['predicted_attack'] = predicted_attack 416 | return df 417 | 418 | #df_attack['kde_val'] = df_attack.apply(lambda row: d[row.aid]['kde'].evaluate(row.time_diffs)[0], axis = 1) 419 | 420 | def alert_by_kde(df,d, p): 421 | """ 422 | Adds column to df that labels at which time_diffs the value of kde is less than the desired kde threshold 423 | """ 424 | df['predicted_attack'] = df.apply(lambda row: row.kde_val <= d[row.aid]['y_thresholds_kde'][p], axis = 1) 425 | return df 426 | 427 | 428 | def alert_by_kde_2(df, dict, p): 429 | """ 430 | Same as alert_by_kde but much slower 431 | """ 432 | predicted_attack = [] 433 | for i in range(df.count()[0]): 434 | aid = df.iloc[i].aid 435 | time_diff = df.iloc[i].time_diffs 436 | if dict[aid]['kde'].evaluate(time_diff) <= dict[aid]['y_thresholds_kde'][p]: 437 | predicted_attack.append(True) 438 | else: 439 | predicted_attack.append(False) 440 | df['predicted_attack'] = predicted_attack 441 | return predicted_attack 442 | 443 | 444 | def alert_by_gauss(df, d, p): 445 | """ 446 | Adds column to df that labels at which time_diffs the value of Gaussian estimate is less than the desired Gaussian threshold 447 | """ 448 | df['predicted_attack'] = df.apply(lambda row: row.gauss_val <= d[row.aid]['y_thresholds_gauss'][p], axis = 1) 449 | return df 450 | 451 | 452 | def alert_by_gauss_2(df, dict, p): 453 | """ 454 | Same as alert_by_gauss but much slower 455 | """ 456 | predicted_attack = [] 457 | for i in range(df.count()[0]): 458 | aid = df.iloc[i].aid 459 | time_diff = df.iloc[i].time_diffs 460 | if dict[aid]['gauss'].pdf(time_diff) <= dict[aid]['y_thresholds_gauss'][p]: 461 | predicted_attack.append(True) 462 | else: 463 | predicted_attack.append(False) 464 | df['predicted_attack'] = predicted_attack 465 | return df 466 | 467 | 468 | def signal_count_new(df, aid): 469 | """ 470 | Calculates how many times a signal with this aid is seen in time window of length mu*4 471 | """ 472 | 473 | # Get the mean inter-arrival time of messages for a specific aid 474 | # times_vec = np.array(df.time[df.aid==aid].to_list()) 475 | times_vec = np.array(df.time[df.PGN==aid].to_list()) 476 | times_vec = times_vec- times_vec.min() 477 | mu = np.diff(times_vec).mean() 478 | 479 | try: 480 | # Get the time breaking points in terms of positions 481 | breakpoints = [bisect.bisect_right(times_vec, i*4*mu) for i in range(int(max(times_vec)/(4*mu)))] 482 | 483 | diffs = list(np.diff(breakpoints)) 484 | for i in range(len(diffs)): 485 | if diffs[i] > 9: 486 | diffs[i] = 9 487 | 488 | # Reflects the distribution od idstance between interarrival times. The limit is 9 489 | count = Counter(np.concatenate([np.arange(0,10), np.array(diffs)])) 490 | return count 491 | except: 492 | return None 493 | 494 | 495 | def prob_dict(df, aid): 496 | """ 497 | Calculates the probability P[i] of seeing a signal with this aid i times in window of length mu*4 498 | Based on calculus and Bayes rules 499 | """ 500 | count_aid = signal_count_new(df, aid) 501 | denom = sum(count_aid[i] for i in range(10)) 502 | P = [0]*10 503 | for i in range(10): 504 | P[i] = count_aid[i]/denom 505 | return P 506 | 507 | 508 | def alert_by_bin(df, d, n=6): 509 | """ 510 | Checks for time windows of length mu*4 (where mu is average time_diff for aid) with 6 or more signals 511 | """ 512 | pd.options.mode.chained_assignment = None 513 | 514 | cm = np.array([[0,0], [0,0]]) 515 | for aid in df.aid.unique(): 516 | #if d[aid]['std'] <= 0.01: 517 | df_test = df[df.aid==aid] 518 | df_test['predicted_attack'] = df_test.time_diffs.rolling(n).sum() <= d[aid]['mu']*4 519 | cm_aid = confusion_matrix(df_test['actual_attack'], df_test['predicted_attack'], labels = [0,1]) 520 | cm += cm_aid 521 | #print(aid, cm_aid) 522 | return cm 523 | 524 | def alert_by_bin_various_p(df, d, p, n=6): 525 | """ 526 | Checks for time windows of length mu*4 (where mu is average time_diff for aid) with 6 or more signals 527 | """ 528 | pd.options.mode.chained_assignment = None 529 | 530 | cm = np.array([[0,0], [0,0]]) 531 | for aid in df.aid.unique(): 532 | #if d[aid]['std'] <= 0.01: 533 | df_test = df[df.aid==aid] 534 | df_test['predicted_attack'] = df_test.time_diffs.rolling(n).sum() <= p*d[aid]['mu'] 535 | cm_aid = confusion_matrix(df_test['actual_attack'], df_test['predicted_attack'], labels = [0,1]) 536 | cm += cm_aid 537 | #print(aid, cm_aid) 538 | return cm 539 | 540 | 541 | def alert_by_bin_J1939(df, d, n=6): 542 | """ 543 | Checks for time windows of length mu*4 (where mu is average time_diff for aid) with 6 or more signals 544 | """ 545 | cm = np.array([[0,0], [0,0]]) 546 | for pgn in df["PGN"].unique(): 547 | try : 548 | if d[pgn]['std'] <= 0.01: 549 | df_test = df[df["PGN"] == pgn] 550 | df_test['predicted_attack'] = df_test.time_diffs.rolling(n).sum() <= d[pgn]['mu']*4 551 | cm_aid = confusion_matrix(df_test['actual_attack'], df_test['predicted_attack'], labels = [0,1]) 552 | cm += cm_aid 553 | #print(aid, cm_aid) 554 | except: 555 | pass 556 | return cm 557 | 558 | 559 | def get_results_mean(attack_list, d): 560 | """ 561 | Marks as attack when 3 or more out of last 6 time_diffs are less than half the mean 562 | Calulates confusion matrix, precision, recall, false positive rate and saves to dictionary 563 | """ 564 | ## Initialize dictionary for results 565 | results_mean = {} 566 | for i in range(len(attack_list)): 567 | results_mean[i+1] = {'cm': [0], 'recall': 0, 'prec':0, 'false_pos':0} 568 | results_mean['total'] = {'cm': [0], 'recall': 0, 'prec':0, 'false_pos':0} 569 | 570 | for i in tqdm(range(len(attack_list))): 571 | attack_list[i] = alert_by_mean(attack_list[i], d) 572 | attack_list[i]['alert_window'] = attack_list[i].predicted_attack.rolling(6).sum() >= 3 573 | cm = confusion_matrix(attack_list[i]['actual_attack'], attack_list[i]['alert_window']) 574 | results_mean[i+1]['cm'] = cm 575 | results_mean[i+1]['prec'] = cm[1,1]/(cm[1,1]+cm[0,1]) 576 | results_mean[i+1]['recall'] = cm[1,1]/(cm[1,1]+cm[1,0]) 577 | results_mean[i+1]['false_pos'] = cm[0,1]/(cm[0,1] + cm[0,0]) 578 | results_mean['total']['cm'] += cm 579 | results_mean['total']['prec'] = results_mean['total']['cm'][1,1]/(results_mean['total']['cm'][1,1]+results_mean['total']['cm'][0,1]) 580 | results_mean['total']['recall'] = results_mean['total']['cm'][1,1]/(results_mean['total']['cm'][1,1]+results_mean['total']['cm'][1,0]) 581 | results_mean['total']['f1'] = 2*((results_mean['total']['prec']*results_mean['total']['recall'])/(results_mean['total']['prec']+results_mean['total']['recall'])) 582 | results_mean['total']['false_pos'] = results_mean['total']['cm'][0,1]/(results_mean['total']['cm'][0,1]+results_mean['total']['cm'][0,0]) 583 | 584 | # print(os.path.dirname(os.getcwd())) 585 | 586 | # return(results_mean) 587 | 588 | picklify(results_mean, os.path.dirname(os.getcwd()) + "/results_mean_final.pkl") 589 | 590 | 591 | def get_results_mean_various_p(attack_list, d): 592 | """ 593 | Marks as attack when 3 or more out of last 6 time_diffs are less than half the mean 594 | Calulates confusion matrix, precision, recall, false positive rate and saves to dictionary 595 | """ 596 | 597 | pvals_mean = np.linspace(0, 1, 19) 598 | 599 | results_mean_final = {} 600 | for p in pvals_mean: 601 | results_mean_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 602 | 603 | for p in tqdm(pvals_mean): 604 | details = np.array([[0,0], [0,0]]) 605 | for i in range(len(attack_list)): 606 | attack_list[i] = alert_by_mean_various_p(attack_list[i], d, p) 607 | attack_list[i]['alert_window'] = attack_list[i].predicted_attack.rolling(6).sum() >= 3 608 | details += confusion_matrix(attack_list[i]['actual_attack'], attack_list[i]['alert_window']) 609 | results_mean_final[p]['cm'] = details 610 | results_mean_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 611 | results_mean_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 612 | results_mean_final[p]['f1'] = 2*((results_mean_final[p]['prec']*results_mean_final[p]['recall'])/(results_mean_final[p]['prec']+results_mean_final[p]['recall'])) 613 | results_mean_final[p]['false_pos'] = details[0,1]/(details[0,1] + details[0,0]) 614 | 615 | # print(os.path.dirname(os.getcwd())) 616 | 617 | # return(results_mean_final) 618 | 619 | picklify(results_mean_final, os.path.dirname(os.getcwd()) + "/results_mean_final.pkl") 620 | 621 | 622 | def get_results_mean_various_p_no_pretraining(attack_list, d): 623 | """ 624 | Marks as attack when 3 or more out of last 6 time_diffs are less than half the mean 625 | Calulates confusion matrix, precision, recall, false positive rate and saves to dictionary 626 | """ 627 | 628 | pvals_mean = np.linspace(0, 1, 19) 629 | 630 | results_mean_final = {} 631 | for p in pvals_mean: 632 | results_mean_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 633 | 634 | for p in tqdm(pvals_mean): 635 | details = np.array([[0,0], [0,0]]) 636 | for i in range(len(attack_list)): 637 | attack_list[i] = alert_by_mean_various_p(attack_list[i], d, p) 638 | attack_list[i]['alert_window'] = attack_list[i].predicted_attack.rolling(6).sum() >= 3 639 | details += confusion_matrix(attack_list[i]['actual_attack'], attack_list[i]['alert_window']) 640 | results_mean_final[p]['cm'] = details 641 | results_mean_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 642 | results_mean_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 643 | results_mean_final[p]['f1'] = 2*((results_mean_final[p]['prec']*results_mean_final[p]['recall'])/(results_mean_final[p]['prec']+results_mean_final[p]['recall'])) 644 | results_mean_final[p]['false_pos'] = details[0,1]/(details[0,1] + details[0,0]) 645 | 646 | # print(os.path.dirname(os.getcwd())) 647 | 648 | # return(results_mean_final) 649 | 650 | picklify(results_mean_final, os.path.dirname(os.getcwd()) + "/results_mean_final_no_pretraining.pkl") 651 | 652 | 653 | def get_results_kde(pvals, attack_list, d): 654 | """ 655 | Marks as attack when last three time_diffs had had p-value less than the p-value threshold for kde 656 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 657 | """ 658 | 659 | pvals_kde = sorted(list(np.arange(0.001, 0.01, 0.001)) + list(np.arange(0, 0.1, 0.01))) 660 | 661 | results_kde_final = {} 662 | for p in pvals_kde: 663 | results_kde_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 664 | 665 | for p in tqdm(pvals): 666 | details = np.array([[0,0], [0,0]]) 667 | for i in range(len(attack_list)): 668 | attack_list[i] = alert_by_kde(attack_list[i], d, p) 669 | attack_list[i]['alert_window'] = attack_list[i].predicted_attack.rolling(3).sum() == 3 670 | details += confusion_matrix(attack_list[i]['actual_attack'], attack_list[i]['alert_window']) 671 | results_kde_final[p]['cm'] = details 672 | results_kde_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 673 | results_kde_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 674 | results_kde_final[p]['f1'] = 2*((results_kde_final[p]['prec']*results_kde_final[p]['recall'])/(results_kde_final[p]['prec']+results_kde_final[p]['recall'])) 675 | results_kde_final[p]['false_pos'] = details[0,1]/(details[0,1]+details[0,0]) 676 | picklify(results_kde_final, os.path.dirname(os.getcwd()) + "/results_kde_final.pkl") 677 | 678 | 679 | def get_results_kde_no_pretraining(pvals, attack_list, d): 680 | """ 681 | Marks as attack when last three time_diffs had had p-value less than the p-value threshold for kde 682 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 683 | """ 684 | 685 | pvals_kde = sorted(list(np.arange(0.001, 0.01, 0.001)) + list(np.arange(0, 0.1, 0.01))) 686 | 687 | results_kde_final = {} 688 | for p in pvals_kde: 689 | results_kde_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 690 | 691 | for p in tqdm(pvals): 692 | details = np.array([[0,0], [0,0]]) 693 | for i in range(len(attack_list)): 694 | attack_list[i] = alert_by_kde(attack_list[i], d, p) 695 | attack_list[i]['alert_window'] = attack_list[i].predicted_attack.rolling(3).sum() == 3 696 | details += confusion_matrix(attack_list[i]['actual_attack'], attack_list[i]['alert_window']) 697 | results_kde_final[p]['cm'] = details 698 | results_kde_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 699 | results_kde_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 700 | results_kde_final[p]['f1'] = 2*((results_kde_final[p]['prec']*results_kde_final[p]['recall'])/(results_kde_final[p]['prec']+results_kde_final[p]['recall'])) 701 | results_kde_final[p]['false_pos'] = details[0,1]/(details[0,1]+details[0,0]) 702 | 703 | picklify(results_kde_final, os.path.dirname(os.getcwd()) + "/results_kde_final_no_pretraining.pkl") 704 | 705 | 706 | def get_results_gauss(attack_list, d): 707 | """ 708 | Marks as attack when last three time_diffs had had p-value less than the p-value threshold for Gaussian distribution 709 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 710 | """ 711 | 712 | pvals_gauss = sorted(list(np.arange(0.001, 0.01, 0.001)) + list(np.arange(0.01, 0.1, 0.01))) 713 | # pvals_gauss = list(np.arange(0.01, 0.21, 0.01)) 714 | # pvals_gauss = list(np.arange(0.0001, 0.0011, 0.0001)) 715 | 716 | results_gauss_final = {} 717 | for p in pvals_gauss: 718 | results_gauss_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 719 | 720 | for p in tqdm(pvals_gauss): 721 | details = np.array([[0,0], [0,0]]) 722 | for i in range(len(attack_list)): 723 | attack_list[i] = alert_by_gauss(attack_list[i], d, p) 724 | attack_list[i]['alert_window'] = attack_list[i].predicted_attack.rolling(3).sum() == 3 725 | details += confusion_matrix(attack_list[i]['actual_attack'], attack_list[i]['alert_window']) 726 | 727 | #print("p: ", p) 728 | print(details) 729 | results_gauss_final[p]['cm'] = details 730 | results_gauss_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 731 | results_gauss_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 732 | results_gauss_final[p]['f1'] = 2*((results_gauss_final[p]['prec']*results_gauss_final[p]['recall'])/(results_gauss_final[p]['prec']+results_gauss_final[p]['recall'])) 733 | results_gauss_final[p]['false_pos'] = details[0,1]/(details[0,1]+details[0,0]) 734 | 735 | picklify(results_gauss_final, os.path.dirname(os.getcwd()) + "/results_gauss_final.pkl") 736 | 737 | 738 | def get_results_gauss_no_pretraining(attack_list, d): 739 | """ 740 | Marks as attack when last three time_diffs had had p-value less than the p-value threshold for Gaussian distribution 741 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 742 | """ 743 | 744 | pvals_gauss = sorted(list(np.arange(0.001, 0.01, 0.001)) + list(np.arange(0.01, 0.1, 0.01))) 745 | # pvals_gauss = list(np.arange(0.01, 0.21, 0.01)) 746 | # pvals_gauss = list(np.arange(0.0001, 0.0011, 0.0001)) 747 | 748 | results_gauss_final = {} 749 | for p in pvals_gauss: 750 | results_gauss_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 751 | 752 | for p in tqdm(pvals_gauss): 753 | details = np.array([[0,0], [0,0]]) 754 | for i in range(len(attack_list)): 755 | attack_list[i] = alert_by_gauss(attack_list[i], d, p) 756 | attack_list[i]['alert_window'] = attack_list[i].predicted_attack.rolling(3).sum() == 3 757 | details += confusion_matrix(attack_list[i]['actual_attack'], attack_list[i]['alert_window']) 758 | 759 | #print("p: ", p) 760 | print(details) 761 | results_gauss_final[p]['cm'] = details 762 | results_gauss_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 763 | results_gauss_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 764 | results_gauss_final[p]['f1'] = 2*((results_gauss_final[p]['prec']*results_gauss_final[p]['recall'])/(results_gauss_final[p]['prec']+results_gauss_final[p]['recall'])) 765 | results_gauss_final[p]['false_pos'] = details[0,1]/(details[0,1]+details[0,0]) 766 | 767 | picklify(results_gauss_final, os.path.dirname(os.getcwd()) + "/results_gauss_final_no_pretraining.pkl") 768 | 769 | 770 | def get_results_binning(attack_list, D, n=6): 771 | """ 772 | Marks as attack when 6 messages with same aid come in less than mu*4 seconds (where mu is average time_diff for the aid) 773 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 774 | """ 775 | 776 | ## Initialize results dictionary 777 | results_binning = {} 778 | for i in range(len(attack_list)): 779 | results_binning[i+1] = {'cm': [0], 'recall': 0, 'prec':0, 'false_pos':0} 780 | results_binning['total'] = {'cm': [0], 'recall': 0, 'prec':0, 'false_pos':0} 781 | 782 | for i in tqdm(range(len(attack_list))): 783 | results_binning[i+1]['cm'] = alert_by_bin(attack_list[i], D, n) 784 | #print(results_binning[i+1]["cm"]) 785 | results_binning[i+1]['prec'] = results_binning[i+1]['cm'][1,1]/(results_binning[i+1]['cm'][1,1]+results_binning[i+1]['cm'][0,1]) 786 | results_binning[i+1]['recall'] = results_binning[i+1]['cm'][1,1]/(results_binning[i+1]['cm'][1,1]+results_binning[i+1]['cm'][1,0]) 787 | results_binning[i+1]['false_pos'] = results_binning[i+1]['cm'][0,1]/(results_binning[i+1]['cm'][0,1]+results_binning[i+1]['cm'][0,0]) 788 | results_binning['total']['cm'] += results_binning[i+1]['cm'] 789 | 790 | #print(results_binning) 791 | results_binning['total']['prec'] = results_binning['total']['cm'][1,1]/(results_binning['total']['cm'][1,1]+results_binning['total']['cm'][0,1]) 792 | results_binning['total']['recall'] = results_binning['total']['cm'][1,1]/(results_binning['total']['cm'][1,1]+results_binning['total']['cm'][1,0]) 793 | results_binning['total']['f1'] = 2*((results_binning['total']['prec']*results_binning['total']['recall'])/(results_binning['total']['prec']+results_binning['total']['recall'])) 794 | results_binning['total']['false_pos'] = results_binning['total']['cm'][0,1]/(results_binning['total']['cm'][0,1]+results_binning['total']['cm'][0,0]) 795 | 796 | #return results_binning 797 | 798 | picklify(results_binning, os.path.dirname(os.getcwd()) + "/results_binning_final.pkl") 799 | 800 | 801 | def get_results_binning_various_p(attack_list, D, n=6): 802 | """ 803 | Marks as attack when 6 messages with same aid come in less than mu*4 seconds (where mu is average time_diff for the aid) 804 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 805 | """ 806 | 807 | pvals_binning = np.linspace(1, 10, 19) # 0, 4 808 | 809 | ## Initialize results dictionary 810 | results_binning_final = {} 811 | for p in pvals_binning: 812 | results_binning_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 813 | 814 | for p in tqdm(pvals_binning): 815 | details = np.array([[0,0], [0,0]]) 816 | for i in range(len(attack_list)): 817 | details += alert_by_bin_various_p(attack_list[i], D, p, n) 818 | results_binning_final[p]["cm"] = details 819 | results_binning_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 820 | results_binning_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 821 | results_binning_final[p]['f1'] = 2*((results_binning_final[p]['prec']*results_binning_final[p]['recall'])/(results_binning_final[p]['prec']+results_binning_final[p]['recall'])) 822 | results_binning_final[p]['false_pos'] = details[0,1]/(details[0,1] + details[0,0]) 823 | 824 | 825 | # return results_binning_final 826 | 827 | picklify(results_binning_final, os.path.dirname(os.getcwd()) + "/results_binning_final.pkl") 828 | 829 | def get_results_binning_various_p_no_pretraining(attack_list, D, n=6): 830 | """ 831 | Marks as attack when 6 messages with same aid come in less than mu*4 seconds (where mu is average time_diff for the aid) 832 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 833 | """ 834 | 835 | pvals_binning = np.linspace(1, 10, 19) # 0, 4 836 | 837 | ## Initialize results dictionary 838 | results_binning_final = {} 839 | for p in pvals_binning: 840 | results_binning_final[p] = {'cm': [0], 'recall': 0, 'prec': 0, 'false_pos': 0} 841 | 842 | for p in tqdm(pvals_binning): 843 | details = np.array([[0,0], [0,0]]) 844 | for i in range(len(attack_list)): 845 | details += alert_by_bin_various_p(attack_list[i], D, p, n) 846 | results_binning_final[p]["cm"] = details 847 | results_binning_final[p]['prec'] = details[1,1]/(details[1,1]+details[0,1]) 848 | results_binning_final[p]['recall'] = details[1,1]/(details[1,1]+details[1,0]) 849 | results_binning_final[p]['f1'] = 2*((results_binning_final[p]['prec']*results_binning_final[p]['recall'])/(results_binning_final[p]['prec']+results_binning_final[p]['recall'])) 850 | results_binning_final[p]['false_pos'] = details[0,1]/(details[0,1] + details[0,0]) 851 | 852 | 853 | # return results_binning_final 854 | 855 | picklify(results_binning_final, os.path.dirname(os.getcwd()) + "/results_binning_final_no_pretraining.pkl") 856 | 857 | 858 | def get_results_binning_J1939(attack_list, D, n=6): 859 | """ 860 | Marks as attack when 6 messages with same aid come in less than mu*4 seconds (where mu is average time_diff for the aid) 861 | Calculates confusion matrix, precision, recall, false positive rate and saves to dictionary 862 | """ 863 | 864 | ## Initialize results dictionary 865 | results_binning = {} 866 | for i in range(len(attack_list)): 867 | results_binning[i+1] = {'cm': [0], 'recall': 0, 'prec':0, 'false_pos':0} 868 | results_binning['total'] = {'cm': [0], 'recall': 0, 'prec':0, 'false_pos':0} 869 | 870 | for i in range(len(attack_list)): 871 | try: 872 | results_binning[i+1]['cm'] = alert_by_bin_J1939(attack_list[i], D, n) 873 | results_binning[i+1]['prec'] = results_binning[i+1]['cm'][1,1]/(results_binning[i+1]['cm'][1,1]+results_binning[i+1]['cm'][0,1]) 874 | results_binning[i+1]['recall'] = results_binning[i+1]['cm'][1,1]/(results_binning[i+1]['cm'][1,1]+results_binning[i+1]['cm'][1,0]) 875 | results_binning[i+1]['false_pos'] = results_binning[i+1]['cm'][0,1]/(results_binning[i+1]['cm'][0,1]+results_binning[i+1]['cm'][0,0]) 876 | results_binning['total']['cm'] += results_binning[i+1]['cm'] 877 | except: 878 | pass 879 | try: 880 | results_binning['total']['prec'] = results_binning['total']['cm'][1,1]/(results_binning['total']['cm'][1,1]+results_binning['total']['cm'][0,1]) 881 | results_binning['total']['recall'] = results_binning['total']['cm'][1,1]/(results_binning['total']['cm'][1,1]+results_binning['total']['cm'][1,0]) 882 | results_binning['total']['false_pos'] = results_binning['total']['cm'][0,1]/(results_binning['total']['cm'][0,1]+results_binning['total']['cm'][0,0]) 883 | except: 884 | pass 885 | return results_binning 886 | 887 | def print_details(y_true, y_hat, filename=''): 888 | cm = confusion_matrix(y_true, y_hat) 889 | prec, rec, fscore, _ = precision_recall_fscore_support( 890 | y_true, y_hat, average='binary', pos_label=1) 891 | 892 | if filename != '': 893 | with open(filename + "_" + ".txt", 'a+') as file: # Use file to refer to the file object 894 | 895 | file.write("Confusion Matrix of %s is \n%r\n" % (name, cm)) 896 | file.write( 897 | f"Prec = {prec:.4f}, recall= {rec:.4f}, fscore = {fscore:.4f}\n\n") 898 | else: 899 | print("Confusion Matrix is \n%r\n" % (cm)) 900 | print( 901 | f"Prec = {prec:.4f}, recall= {rec:.4f}, fscore = {fscore:.4f}\n\n") 902 | return {'cm': cm, 'prec': prec, 'recall': rec, 'fscore': fscore} 903 | -------------------------------------------------------------------------------- /figs/figure_prc.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/figure_prc.pdf -------------------------------------------------------------------------------- /figs/figure_prc_condensed.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/figure_prc_condensed.pdf -------------------------------------------------------------------------------- /figs/figure_prc_no_pretraining.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/figure_prc_no_pretraining.pdf -------------------------------------------------------------------------------- /figs/testing_distributions.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/testing_distributions.pdf -------------------------------------------------------------------------------- /figs/training_distributions.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/training_distributions.pdf -------------------------------------------------------------------------------- /figs/training_distributions_1760.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/training_distributions_1760.pdf -------------------------------------------------------------------------------- /figs/training_distributions_1760_fit.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/training_distributions_1760_fit.pdf -------------------------------------------------------------------------------- /figs/training_distributions_208.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/training_distributions_208.pdf -------------------------------------------------------------------------------- /figs/training_distributions_208_fit.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/training_distributions_208_fit.pdf -------------------------------------------------------------------------------- /figs/training_distributions_fit_combined.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/pmoriano/can-time-based-ids-benchmark/e0c862d6c6cbb0d60ed2572a90dd483a3253ed0f/figs/training_distributions_fit_combined.pdf --------------------------------------------------------------------------------