sibe (empty) → 0.1.0.0
raw patch · 14 files changed
+1928/−0 lines, 14 filesdep +Chartdep +Chart-cairodep +JuicyPixelssetup-changed
Dependencies added: Chart, Chart-cairo, JuicyPixels, base, containers, data-default-class, deepseq, directory, hmatrix, lens, random, random-shuffle, regex-base, regex-pcre, sibe, split, stemmer, text, vector
Files
- LICENSE +674/−0
- Setup.hs +2/−0
- examples/424encoder.hs +49/−0
- examples/naivebayes-doc-classifier.hs +82/−0
- examples/notmnist.hs +103/−0
- examples/word2vec.hs +88/−0
- examples/xor.hs +40/−0
- sibe.cabal +111/−0
- src/Sibe.hs +362/−0
- src/Sibe/NLP.hs +122/−0
- src/Sibe/NaiveBayes.hs +128/−0
- src/Sibe/Utils.hs +28/−0
- src/Sibe/Word2Vec.hs +131/−0
- test/Spec.hs +8/−0
+ LICENSE view
@@ -0,0 +1,674 @@+ GNU GENERAL PUBLIC LICENSE+ Version 3, 29 June 2007++ Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>+ Everyone is permitted to copy and distribute verbatim copies+ of this license document, but changing it is not allowed.++ Preamble++ The GNU General Public License is a free, copyleft license for+software and other kinds of works.++ The licenses for most software and other practical works are designed+to take away your freedom to share and change the works. By contrast,+the GNU General Public License is intended to guarantee your freedom to+share and change all versions of a program--to make sure it remains free+software for all its users. We, the Free Software Foundation, use the+GNU General Public License for most of our software; it applies also to+any other work released this way by its authors. You can apply it to+your programs, too.++ When we speak of free software, we are referring to freedom, not+price. Our General Public Licenses are designed to make sure that you+have the freedom to distribute copies of free software (and charge for+them if you wish), that you receive source code or can get it if you+want it, that you can change the software or use pieces of it in new+free programs, and that you know you can do these things.++ To protect your rights, we need to prevent others from denying you+these rights or asking you to surrender the rights. Therefore, you have+certain responsibilities if you distribute copies of the software, or if+you modify it: responsibilities to respect the freedom of others.++ For example, if you distribute copies of such a program, whether+gratis or for a fee, you must pass on to the recipients the same+freedoms that you received. You must make sure that they, too, receive+or can get the source code. And you must show them these terms so they+know their rights.++ Developers that use the GNU GPL protect your rights with two steps:+(1) assert copyright on the software, and (2) offer you this License+giving you legal permission to copy, distribute and/or modify it.++ For the developers' and authors' protection, the GPL clearly explains+that there is no warranty for this free software. For both users' and+authors' sake, the GPL requires that modified versions be marked as+changed, so that their problems will not be attributed erroneously to+authors of previous versions.++ Some devices are designed to deny users access to install or run+modified versions of the software inside them, although the manufacturer+can do so. This is fundamentally incompatible with the aim of+protecting users' freedom to change the software. The systematic+pattern of such abuse occurs in the area of products for individuals to+use, which is precisely where it is most unacceptable. Therefore, we+have designed this version of the GPL to prohibit the practice for those+products. If such problems arise substantially in other domains, we+stand ready to extend this provision to those domains in future versions+of the GPL, as needed to protect the freedom of users.++ Finally, every program is threatened constantly by software patents.+States should not allow patents to restrict development and use of+software on general-purpose computers, but in those that do, we wish to+avoid the special danger that patents applied to a free program could+make it effectively proprietary. To prevent this, the GPL assures that+patents cannot be used to render the program non-free.++ The precise terms and conditions for copying, distribution and+modification follow.++ TERMS AND CONDITIONS++ 0. Definitions.++ "This License" refers to version 3 of the GNU General Public License.++ "Copyright" also means copyright-like laws that apply to other kinds of+works, such as semiconductor masks.++ "The Program" refers to any copyrightable work licensed under this+License. Each licensee is addressed as "you". "Licensees" and+"recipients" may be individuals or organizations.++ To "modify" a work means to copy from or adapt all or part of the work+in a fashion requiring copyright permission, other than the making of an+exact copy. The resulting work is called a "modified version" of the+earlier work or a work "based on" the earlier work.++ A "covered work" means either the unmodified Program or a work based+on the Program.++ To "propagate" a work means to do anything with it that, without+permission, would make you directly or secondarily liable for+infringement under applicable copyright law, except executing it on a+computer or modifying a private copy. Propagation includes copying,+distribution (with or without modification), making available to the+public, and in some countries other activities as well.++ To "convey" a work means any kind of propagation that enables other+parties to make or receive copies. Mere interaction with a user through+a computer network, with no transfer of a copy, is not conveying.++ An interactive user interface displays "Appropriate Legal Notices"+to the extent that it includes a convenient and prominently visible+feature that (1) displays an appropriate copyright notice, and (2)+tells the user that there is no warranty for the work (except to the+extent that warranties are provided), that licensees may convey the+work under this License, and how to view a copy of this License. If+the interface presents a list of user commands or options, such as a+menu, a prominent item in the list meets this criterion.++ 1. Source Code.++ The "source code" for a work means the preferred form of the work+for making modifications to it. "Object code" means any non-source+form of a work.++ A "Standard Interface" means an interface that either is an official+standard defined by a recognized standards body, or, in the case of+interfaces specified for a particular programming language, one that+is widely used among developers working in that language.++ The "System Libraries" of an executable work include anything, other+than the work as a whole, that (a) is included in the normal form of+packaging a Major Component, but which is not part of that Major+Component, and (b) serves only to enable use of the work with that+Major Component, or to implement a Standard Interface for which an+implementation is available to the public in source code form. A+"Major Component", in this context, means a major essential component+(kernel, window system, and so on) of the specific operating system+(if any) on which the executable work runs, or a compiler used to+produce the work, or an object code interpreter used to run it.++ The "Corresponding Source" for a work in object code form means all+the source code needed to generate, install, and (for an executable+work) run the object code and to modify the work, including scripts to+control those activities. However, it does not include the work's+System Libraries, or general-purpose tools or generally available free+programs which are used unmodified in performing those activities but+which are not part of the work. For example, Corresponding Source+includes interface definition files associated with source files for+the work, and the source code for shared libraries and dynamically+linked subprograms that the work is specifically designed to require,+such as by intimate data communication or control flow between those+subprograms and other parts of the work.++ The Corresponding Source need not include anything that users+can regenerate automatically from other parts of the Corresponding+Source.++ The Corresponding Source for a work in source code form is that+same work.++ 2. Basic Permissions.++ All rights granted under this License are granted for the term of+copyright on the Program, and are irrevocable provided the stated+conditions are met. This License explicitly affirms your unlimited+permission to run the unmodified Program. The output from running a+covered work is covered by this License only if the output, given its+content, constitutes a covered work. This License acknowledges your+rights of fair use or other equivalent, as provided by copyright law.++ You may make, run and propagate covered works that you do not+convey, without conditions so long as your license otherwise remains+in force. You may convey covered works to others for the sole purpose+of having them make modifications exclusively for you, or provide you+with facilities for running those works, provided that you comply with+the terms of this License in conveying all material for which you do+not control copyright. Those thus making or running the covered works+for you must do so exclusively on your behalf, under your direction+and control, on terms that prohibit them from making any copies of+your copyrighted material outside their relationship with you.++ Conveying under any other circumstances is permitted solely under+the conditions stated below. Sublicensing is not allowed; section 10+makes it unnecessary.++ 3. Protecting Users' Legal Rights From Anti-Circumvention Law.++ No covered work shall be deemed part of an effective technological+measure under any applicable law fulfilling obligations under article+11 of the WIPO copyright treaty adopted on 20 December 1996, or+similar laws prohibiting or restricting circumvention of such+measures.++ When you convey a covered work, you waive any legal power to forbid+circumvention of technological measures to the extent such circumvention+is effected by exercising rights under this License with respect to+the covered work, and you disclaim any intention to limit operation or+modification of the work as a means of enforcing, against the work's+users, your or third parties' legal rights to forbid circumvention of+technological measures.++ 4. Conveying Verbatim Copies.++ You may convey verbatim copies of the Program's source code as you+receive it, in any medium, provided that you conspicuously and+appropriately publish on each copy an appropriate copyright notice;+keep intact all notices stating that this License and any+non-permissive terms added in accord with section 7 apply to the code;+keep intact all notices of the absence of any warranty; and give all+recipients a copy of this License along with the Program.++ You may charge any price or no price for each copy that you convey,+and you may offer support or warranty protection for a fee.++ 5. Conveying Modified Source Versions.++ You may convey a work based on the Program, or the modifications to+produce it from the Program, in the form of source code under the+terms of section 4, provided that you also meet all of these conditions:++ a) The work must carry prominent notices stating that you modified+ it, and giving a relevant date.++ b) The work must carry prominent notices stating that it is+ released under this License and any conditions added under section+ 7. This requirement modifies the requirement in section 4 to+ "keep intact all notices".++ c) You must license the entire work, as a whole, under this+ License to anyone who comes into possession of a copy. This+ License will therefore apply, along with any applicable section 7+ additional terms, to the whole of the work, and all its parts,+ regardless of how they are packaged. This License gives no+ permission to license the work in any other way, but it does not+ invalidate such permission if you have separately received it.++ d) If the work has interactive user interfaces, each must display+ Appropriate Legal Notices; however, if the Program has interactive+ interfaces that do not display Appropriate Legal Notices, your+ work need not make them do so.++ A compilation of a covered work with other separate and independent+works, which are not by their nature extensions of the covered work,+and which are not combined with it such as to form a larger program,+in or on a volume of a storage or distribution medium, is called an+"aggregate" if the compilation and its resulting copyright are not+used to limit the access or legal rights of the compilation's users+beyond what the individual works permit. Inclusion of a covered work+in an aggregate does not cause this License to apply to the other+parts of the aggregate.++ 6. Conveying Non-Source Forms.++ You may convey a covered work in object code form under the terms+of sections 4 and 5, provided that you also convey the+machine-readable Corresponding Source under the terms of this License,+in one of these ways:++ a) Convey the object code in, or embodied in, a physical product+ (including a physical distribution medium), accompanied by the+ Corresponding Source fixed on a durable physical medium+ customarily used for software interchange.++ b) Convey the object code in, or embodied in, a physical product+ (including a physical distribution medium), accompanied by a+ written offer, valid for at least three years and valid for as+ long as you offer spare parts or customer support for that product+ model, to give anyone who possesses the object code either (1) a+ copy of the Corresponding Source for all the software in the+ product that is covered by this License, on a durable physical+ medium customarily used for software interchange, for a price no+ more than your reasonable cost of physically performing this+ conveying of source, or (2) access to copy the+ Corresponding Source from a network server at no charge.++ c) Convey individual copies of the object code with a copy of the+ written offer to provide the Corresponding Source. This+ alternative is allowed only occasionally and noncommercially, and+ only if you received the object code with such an offer, in accord+ with subsection 6b.++ d) Convey the object code by offering access from a designated+ place (gratis or for a charge), and offer equivalent access to the+ Corresponding Source in the same way through the same place at no+ further charge. You need not require recipients to copy the+ Corresponding Source along with the object code. If the place to+ copy the object code is a network server, the Corresponding Source+ may be on a different server (operated by you or a third party)+ that supports equivalent copying facilities, provided you maintain+ clear directions next to the object code saying where to find the+ Corresponding Source. Regardless of what server hosts the+ Corresponding Source, you remain obligated to ensure that it is+ available for as long as needed to satisfy these requirements.++ e) Convey the object code using peer-to-peer transmission, provided+ you inform other peers where the object code and Corresponding+ Source of the work are being offered to the general public at no+ charge under subsection 6d.++ A separable portion of the object code, whose source code is excluded+from the Corresponding Source as a System Library, need not be+included in conveying the object code work.++ A "User Product" is either (1) a "consumer product", which means any+tangible personal property which is normally used for personal, family,+or household purposes, or (2) anything designed or sold for incorporation+into a dwelling. In determining whether a product is a consumer product,+doubtful cases shall be resolved in favor of coverage. For a particular+product received by a particular user, "normally used" refers to a+typical or common use of that class of product, regardless of the status+of the particular user or of the way in which the particular user+actually uses, or expects or is expected to use, the product. A product+is a consumer product regardless of whether the product has substantial+commercial, industrial or non-consumer uses, unless such uses represent+the only significant mode of use of the product.++ "Installation Information" for a User Product means any methods,+procedures, authorization keys, or other information required to install+and execute modified versions of a covered work in that User Product from+a modified version of its Corresponding Source. The information must+suffice to ensure that the continued functioning of the modified object+code is in no case prevented or interfered with solely because+modification has been made.++ If you convey an object code work under this section in, or with, or+specifically for use in, a User Product, and the conveying occurs as+part of a transaction in which the right of possession and use of the+User Product is transferred to the recipient in perpetuity or for a+fixed term (regardless of how the transaction is characterized), the+Corresponding Source conveyed under this section must be accompanied+by the Installation Information. But this requirement does not apply+if neither you nor any third party retains the ability to install+modified object code on the User Product (for example, the work has+been installed in ROM).++ The requirement to provide Installation Information does not include a+requirement to continue to provide support service, warranty, or updates+for a work that has been modified or installed by the recipient, or for+the User Product in which it has been modified or installed. Access to a+network may be denied when the modification itself materially and+adversely affects the operation of the network or violates the rules and+protocols for communication across the network.++ Corresponding Source conveyed, and Installation Information provided,+in accord with this section must be in a format that is publicly+documented (and with an implementation available to the public in+source code form), and must require no special password or key for+unpacking, reading or copying.++ 7. Additional Terms.++ "Additional permissions" are terms that supplement the terms of this+License by making exceptions from one or more of its conditions.+Additional permissions that are applicable to the entire Program shall+be treated as though they were included in this License, to the extent+that they are valid under applicable law. If additional permissions+apply only to part of the Program, that part may be used separately+under those permissions, but the entire Program remains governed by+this License without regard to the additional permissions.++ When you convey a copy of a covered work, you may at your option+remove any additional permissions from that copy, or from any part of+it. (Additional permissions may be written to require their own+removal in certain cases when you modify the work.) You may place+additional permissions on material, added by you to a covered work,+for which you have or can give appropriate copyright permission.++ Notwithstanding any other provision of this License, for material you+add to a covered work, you may (if authorized by the copyright holders of+that material) supplement the terms of this License with terms:++ a) Disclaiming warranty or limiting liability differently from the+ terms of sections 15 and 16 of this License; or++ b) Requiring preservation of specified reasonable legal notices or+ author attributions in that material or in the Appropriate Legal+ Notices displayed by works containing it; or++ c) Prohibiting misrepresentation of the origin of that material, or+ requiring that modified versions of such material be marked in+ reasonable ways as different from the original version; or++ d) Limiting the use for publicity purposes of names of licensors or+ authors of the material; or++ e) Declining to grant rights under trademark law for use of some+ trade names, trademarks, or service marks; or++ f) Requiring indemnification of licensors and authors of that+ material by anyone who conveys the material (or modified versions of+ it) with contractual assumptions of liability to the recipient, for+ any liability that these contractual assumptions directly impose on+ those licensors and authors.++ All other non-permissive additional terms are considered "further+restrictions" within the meaning of section 10. If the Program as you+received it, or any part of it, contains a notice stating that it is+governed by this License along with a term that is a further+restriction, you may remove that term. If a license document contains+a further restriction but permits relicensing or conveying under this+License, you may add to a covered work material governed by the terms+of that license document, provided that the further restriction does+not survive such relicensing or conveying.++ If you add terms to a covered work in accord with this section, you+must place, in the relevant source files, a statement of the+additional terms that apply to those files, or a notice indicating+where to find the applicable terms.++ Additional terms, permissive or non-permissive, may be stated in the+form of a separately written license, or stated as exceptions;+the above requirements apply either way.++ 8. Termination.++ You may not propagate or modify a covered work except as expressly+provided under this License. Any attempt otherwise to propagate or+modify it is void, and will automatically terminate your rights under+this License (including any patent licenses granted under the third+paragraph of section 11).++ However, if you cease all violation of this License, then your+license from a particular copyright holder is reinstated (a)+provisionally, unless and until the copyright holder explicitly and+finally terminates your license, and (b) permanently, if the copyright+holder fails to notify you of the violation by some reasonable means+prior to 60 days after the cessation.++ Moreover, your license from a particular copyright holder is+reinstated permanently if the copyright holder notifies you of the+violation by some reasonable means, this is the first time you have+received notice of violation of this License (for any work) from that+copyright holder, and you cure the violation prior to 30 days after+your receipt of the notice.++ Termination of your rights under this section does not terminate the+licenses of parties who have received copies or rights from you under+this License. If your rights have been terminated and not permanently+reinstated, you do not qualify to receive new licenses for the same+material under section 10.++ 9. Acceptance Not Required for Having Copies.++ You are not required to accept this License in order to receive or+run a copy of the Program. Ancillary propagation of a covered work+occurring solely as a consequence of using peer-to-peer transmission+to receive a copy likewise does not require acceptance. However,+nothing other than this License grants you permission to propagate or+modify any covered work. These actions infringe copyright if you do+not accept this License. Therefore, by modifying or propagating a+covered work, you indicate your acceptance of this License to do so.++ 10. Automatic Licensing of Downstream Recipients.++ Each time you convey a covered work, the recipient automatically+receives a license from the original licensors, to run, modify and+propagate that work, subject to this License. You are not responsible+for enforcing compliance by third parties with this License.++ An "entity transaction" is a transaction transferring control of an+organization, or substantially all assets of one, or subdividing an+organization, or merging organizations. If propagation of a covered+work results from an entity transaction, each party to that+transaction who receives a copy of the work also receives whatever+licenses to the work the party's predecessor in interest had or could+give under the previous paragraph, plus a right to possession of the+Corresponding Source of the work from the predecessor in interest, if+the predecessor has it or can get it with reasonable efforts.++ You may not impose any further restrictions on the exercise of the+rights granted or affirmed under this License. For example, you may+not impose a license fee, royalty, or other charge for exercise of+rights granted under this License, and you may not initiate litigation+(including a cross-claim or counterclaim in a lawsuit) alleging that+any patent claim is infringed by making, using, selling, offering for+sale, or importing the Program or any portion of it.++ 11. Patents.++ A "contributor" is a copyright holder who authorizes use under this+License of the Program or a work on which the Program is based. The+work thus licensed is called the contributor's "contributor version".++ A contributor's "essential patent claims" are all patent claims+owned or controlled by the contributor, whether already acquired or+hereafter acquired, that would be infringed by some manner, permitted+by this License, of making, using, or selling its contributor version,+but do not include claims that would be infringed only as a+consequence of further modification of the contributor version. For+purposes of this definition, "control" includes the right to grant+patent sublicenses in a manner consistent with the requirements of+this License.++ Each contributor grants you a non-exclusive, worldwide, royalty-free+patent license under the contributor's essential patent claims, to+make, use, sell, offer for sale, import and otherwise run, modify and+propagate the contents of its contributor version.++ In the following three paragraphs, a "patent license" is any express+agreement or commitment, however denominated, not to enforce a patent+(such as an express permission to practice a patent or covenant not to+sue for patent infringement). To "grant" such a patent license to a+party means to make such an agreement or commitment not to enforce a+patent against the party.++ If you convey a covered work, knowingly relying on a patent license,+and the Corresponding Source of the work is not available for anyone+to copy, free of charge and under the terms of this License, through a+publicly available network server or other readily accessible means,+then you must either (1) cause the Corresponding Source to be so+available, or (2) arrange to deprive yourself of the benefit of the+patent license for this particular work, or (3) arrange, in a manner+consistent with the requirements of this License, to extend the patent+license to downstream recipients. "Knowingly relying" means you have+actual knowledge that, but for the patent license, your conveying the+covered work in a country, or your recipient's use of the covered work+in a country, would infringe one or more identifiable patents in that+country that you have reason to believe are valid.++ If, pursuant to or in connection with a single transaction or+arrangement, you convey, or propagate by procuring conveyance of, a+covered work, and grant a patent license to some of the parties+receiving the covered work authorizing them to use, propagate, modify+or convey a specific copy of the covered work, then the patent license+you grant is automatically extended to all recipients of the covered+work and works based on it.++ A patent license is "discriminatory" if it does not include within+the scope of its coverage, prohibits the exercise of, or is+conditioned on the non-exercise of one or more of the rights that are+specifically granted under this License. You may not convey a covered+work if you are a party to an arrangement with a third party that is+in the business of distributing software, under which you make payment+to the third party based on the extent of your activity of conveying+the work, and under which the third party grants, to any of the+parties who would receive the covered work from you, a discriminatory+patent license (a) in connection with copies of the covered work+conveyed by you (or copies made from those copies), or (b) primarily+for and in connection with specific products or compilations that+contain the covered work, unless you entered into that arrangement,+or that patent license was granted, prior to 28 March 2007.++ Nothing in this License shall be construed as excluding or limiting+any implied license or other defenses to infringement that may+otherwise be available to you under applicable patent law.++ 12. No Surrender of Others' Freedom.++ If conditions are imposed on you (whether by court order, agreement or+otherwise) that contradict the conditions of this License, they do not+excuse you from the conditions of this License. If you cannot convey a+covered work so as to satisfy simultaneously your obligations under this+License and any other pertinent obligations, then as a consequence you may+not convey it at all. For example, if you agree to terms that obligate you+to collect a royalty for further conveying from those to whom you convey+the Program, the only way you could satisfy both those terms and this+License would be to refrain entirely from conveying the Program.++ 13. Use with the GNU Affero General Public License.++ Notwithstanding any other provision of this License, you have+permission to link or combine any covered work with a work licensed+under version 3 of the GNU Affero General Public License into a single+combined work, and to convey the resulting work. The terms of this+License will continue to apply to the part which is the covered work,+but the special requirements of the GNU Affero General Public License,+section 13, concerning interaction through a network will apply to the+combination as such.++ 14. Revised Versions of this License.++ The Free Software Foundation may publish revised and/or new versions of+the GNU General Public License from time to time. Such new versions will+be similar in spirit to the present version, but may differ in detail to+address new problems or concerns.++ Each version is given a distinguishing version number. If the+Program specifies that a certain numbered version of the GNU General+Public License "or any later version" applies to it, you have the+option of following the terms and conditions either of that numbered+version or of any later version published by the Free Software+Foundation. If the Program does not specify a version number of the+GNU General Public License, you may choose any version ever published+by the Free Software Foundation.++ If the Program specifies that a proxy can decide which future+versions of the GNU General Public License can be used, that proxy's+public statement of acceptance of a version permanently authorizes you+to choose that version for the Program.++ Later license versions may give you additional or different+permissions. However, no additional obligations are imposed on any+author or copyright holder as a result of your choosing to follow a+later version.++ 15. Disclaimer of Warranty.++ THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY+APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR+PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM+IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.++ 16. Limitation of Liability.++ IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF+SUCH DAMAGES.++ 17. Interpretation of Sections 15 and 16.++ If the disclaimer of warranty and limitation of liability provided+above cannot be given local legal effect according to their terms,+reviewing courts shall apply local law that most closely approximates+an absolute waiver of all civil liability in connection with the+Program, unless a warranty or assumption of liability accompanies a+copy of the Program in return for a fee.++ END OF TERMS AND CONDITIONS++ How to Apply These Terms to Your New Programs++ If you develop a new program, and you want it to be of the greatest+possible use to the public, the best way to achieve this is to make it+free software which everyone can redistribute and change under these terms.++ To do so, attach the following notices to the program. It is safest+to attach them to the start of each source file to most effectively+state the exclusion of warranty; and each file should have at least+the "copyright" line and a pointer to where the full notice is found.++ {one line to give the program's name and a brief idea of what it does.}+ Copyright (C) {year} {name of author}++ This program is free software: you can redistribute it and/or modify+ it under the terms of the GNU General Public License as published by+ the Free Software Foundation, either version 3 of the License, or+ (at your option) any later version.++ This program is distributed in the hope that it will be useful,+ but WITHOUT ANY WARRANTY; without even the implied warranty of+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the+ GNU General Public License for more details.++ You should have received a copy of the GNU General Public License+ along with this program. If not, see <http://www.gnu.org/licenses/>.++Also add information on how to contact you by electronic and paper mail.++ If the program does terminal interaction, make it output a short+notice like this when it starts in an interactive mode:++ Sibe Copyright (C) 2016 Mahdi Dibaiee+ This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.+ This is free software, and you are welcome to redistribute it+ under certain conditions; type `show c' for details.++The hypothetical commands `show w' and `show c' should show the appropriate+parts of the General Public License. Of course, your program's commands+might be different; for a GUI interface, you would use an "about box".++ You should also get your employer (if you work as a programmer) or school,+if any, to sign a "copyright disclaimer" for the program, if necessary.+For more information on this, and how to apply and follow the GNU GPL, see+<http://www.gnu.org/licenses/>.++ The GNU General Public License does not permit incorporating your program+into proprietary programs. If your program is a subroutine library, you+may consider it more useful to permit linking proprietary applications with+the library. If this is what you want to do, use the GNU Lesser General+Public License instead of this License. But first, please read+<http://www.gnu.org/philosophy/why-not-lgpl.html>.
+ Setup.hs view
@@ -0,0 +1,2 @@+import Distribution.Simple+main = defaultMain
+ examples/424encoder.hs view
@@ -0,0 +1,49 @@+module Main where+ import Sibe+ import Numeric.LinearAlgebra+ import Data.List+ import Debug.Trace+ import Data.Default.Class++ main = do+ let alpha = 0.5+ epochs = 1000+ a = (sigmoid, sigmoid')+ rnetwork = randomNetwork 0 (-0.1, 0.1) 4 [(2, a)] (4, a)++ inputs = [vector [1, 0, 0, 0],+ vector [0, 1, 0, 0],+ vector [0, 0, 1, 0],+ vector [0, 0, 0, 1]]++ labels = [vector [1, 0, 0, 0],+ vector [0, 1, 0, 0],+ vector [0, 0, 1, 0],+ vector [0, 0, 0, 1]]++ session = def { network = rnetwork+ , learningRate = 0.5+ , epochs = 1000+ , training = zip inputs labels+ , test = zip inputs labels+ } :: Session++ let initialCost = crossEntropy session++ newsession <- run gd session++ let results = map (`forward` newsession) inputs+ rounded = map (map round . toList) results++ cost = crossEntropy newsession++ putStrLn "parameters: "+ putStrLn $ "- inputs: " ++ show inputs+ putStrLn $ "- labels: " ++ show labels+ putStrLn $ "- learning rate: " ++ show alpha+ putStrLn $ "- epochs: " ++ show epochs+ putStrLn $ "- initial cost (cross-entropy): " ++ show initialCost+ putStrLn "results: "+ putStrLn $ "- actual result: " ++ show results+ putStrLn $ "- rounded result: " ++ show rounded+ putStrLn $ "- cost (cross-entropy): " ++ show cost
+ examples/naivebayes-doc-classifier.hs view
@@ -0,0 +1,82 @@+module Main+ where+ -- import Sibe+ import Sibe.NLP+ import Sibe.NaiveBayes+ import Text.Printf+ import Data.List+ import Data.Maybe+ import Debug.Trace+ import Data.List.Split+ import Control.Arrow ((&&&))+ import Control.Monad (when, unless)+ import Data.Function (on)+ import System.Environment++ main = do+ args <- getArgs+ dataset <- readFile "examples/doc-classifier-data/data-reuters"+ test <- readFile "examples/doc-classifier-data/data-reuters-test"++ classes <- map (filter (/= ' ')) . lines <$> readFile "examples/doc-classifier-data/data-classes"+ sws <- lines <$> readFile "examples/stopwords"++ let verbose = elem "-v" args || elem "--verbose" args+ topten = elem "-10" args || elem "--top-ten" args+ unless verbose $ putStrLn "use --verbose to print more information"++ let intClasses = [0..length classes - 1]+ documents = cleanDocuments . removeWords sws $ createDocuments classes dataset+ testDocuments = cleanDocuments $ createDocuments classes test++ nb = initialize documents intClasses++ -- top-ten+ topClasses = take 10 . reverse $ sortBy (compare `on` (length . snd)) (cd nb)+ filtered = map (\(c, ds) -> (c, take 100 ds)) topClasses+ filteredClasses = map fst filtered+ ttDocs = concatMap snd filtered+ ttNB = initialize ttDocs filteredClasses++ ttTestDocuments = filter ((`elem` filteredClasses) . c) . cleanDocuments $ createDocuments classes test++ ttResults = session ttTestDocuments ttNB+ normalResults = session testDocuments nb+ results = if topten then ttResults else normalResults++ iClasses = if topten then filteredClasses else intClasses+ -- results = session devTestDocuments nb++ when verbose . putStrLn $ "# Example of cleaned document:\n" ++ (show . text $ head documents)++ let showResults (c, (r, confidence)) = putStrLn (classes !! c ++ " ~ " ++ classes !! r)+ when verbose $ mapM_ showResults results++ when (verbose && not topten) .+ putStrLn $ "The training data is imbalanced which causes the classifier to be biased towards\n"+ ++ "some classes, `earn` is an example, the class alone has around 90% accuracy while\n"+ ++ "the rest of classes have a much lower accuracy and it's commonly seen that most inputs\n"+ ++ "are incorrectly classified as `earn`.\n"+ ++ "Try running with --top-ten to classify top 10 classes by using evenly split documents\n"++ let+ accuracies =+ let as = zip iClasses $ map (\c -> filter ((==c) . fst) results) iClasses+ av = filter (not . null . snd) as+ calculated = map (fst &&& accuracy . snd) av+ in sortBy (\(_, a) (_, b) -> b `compare` a) calculated++ when verbose $+ mapM_ (\(c, a) -> putStrLn $ "Accuracy(" ++ classes !! c ++ ") = " ++ show a) accuracies++ putStrLn $ "\nAverages: "+ putStrLn $ "Recall = " ++ show (recall results)+ putStrLn $ "Precision = " ++ show (precision results)+ putStrLn $ "F Measure = " ++ show (fmeasure results)+ putStrLn $ "Accuracy = " ++ show (accuracy results)++ createDocuments classes content =+ let splitted = splitOn (replicate 10 '-' ++ "\n") content+ pairs = map ((head . lines) &&& (unwords . tail . lines)) splitted+ documents = map (\(topic, text) -> Document text (fromJust $ elemIndex topic classes)) pairs+ in documents
+ examples/notmnist.hs view
@@ -0,0 +1,103 @@+{-# LANGUAGE RecordWildCards #-}+{-# LANGUAGE FlexibleContexts #-}+{-# LANGUAGE ScopedTypeVariables #-}++module Main where+ import Sibe+ import Numeric.LinearAlgebra+ import Data.List+ import Debug.Trace+ import System.IO+ import System.Directory+ import Codec.Picture+ import Codec.Picture.Types+ import qualified Data.Vector.Storable as V+ import Data.Either+ import System.Random+ import System.Random.Shuffle+ import Data.Default.Class++ main = do+ -- random seed, you might comment this line to get real random results+ setStdGen (mkStdGen 100)++ let a = (sigmoid, sigmoid')+ o = (softmax, crossEntropy')+ rnetwork = randomNetwork 0 (-1, 1) (28*28) [(100, a)] (10, o)++ (inputs, labels) <- dataset++ let trp = length inputs * 70 `div` 100+ tep = length inputs * 30 `div` 100++ -- training data+ trinputs = take trp inputs+ trlabels = take trp labels++ -- test data+ teinputs = take tep . drop trp $ inputs+ telabels = take tep . drop trp $ labels++ let session = def { learningRate = 0.5+ , batchSize = 32+ , epochs = 10+ , network = rnetwork+ , training = zip trinputs trlabels+ , test = zip teinputs telabels+ , drawChart = True+ , chartName = "notmnist.png"+ } :: Session++ let initialCost = crossEntropy session++ newsession <- run (sgd . learningRateDecay (1.1, 5e-2)) session++ let cost = crossEntropy newsession++ putStrLn "parameters: "+ putStrLn $ "- batch size: " ++ show (batchSize session)+ putStrLn $ "- learning rate: " ++ show (learningRate session)+ putStrLn $ "- epochs: " ++ show (epochs session)+ putStrLn $ "- initial cost (cross-entropy): " ++ show initialCost+ putStrLn "results: "+ putStrLn $ "- accuracy: " ++ show (accuracy newsession)+ putStrLn $ "- cost (cross-entropy): " ++ show cost++ dataset :: IO ([Vector Double], [Vector Double])+ dataset = do+ let dir = "examples/notMNIST/"+ + groups <- filter ((/= '.') . head) <$> listDirectory dir++ inputFiles <- mapM (listDirectory . (dir ++)) groups++ let n = 512 {-- minimum (map length inputFiles) --}+ numbers = map (`div` n) [0..n * length groups - 1]+ inputFilesFull = map (\(i, g) -> map ((dir ++ i ++ "/") ++) g) (zip groups inputFiles)+++ inputImages <- mapM (mapM readImage . take n) inputFilesFull++ let names = map (take n) inputFilesFull++ let (l, r) = partitionEithers $ concat inputImages+ inputs = map (fromPixels . convertRGB8) r+ labels = map (\i -> V.replicate i 0 `V.snoc` 1 V.++ V.replicate (9 - i) 0) numbers++ pairs = zip inputs labels++ shuffled <- shuffleM pairs+ return (map fst shuffled, map snd shuffled)++ where+ fromPixels :: Image PixelRGB8 -> Vector Double+ fromPixels img@Image { .. } =+ let pairs = [(x, y) | x <- [0..imageWidth - 1], y <- [0..imageHeight - 1]]+ in V.fromList $ map iter pairs+ where+ iter (x, y) =+ let (PixelRGB8 r g b) = convertPixel $ pixelAt img x y+ in+ if r == 0 && g == 0 && b == 0 then 0 else 1++
+ examples/word2vec.hs view
@@ -0,0 +1,88 @@+{-# LANGUAGE RecordWildCards #-}+{-# LANGUAGE FlexibleContexts #-}+{-# LANGUAGE ScopedTypeVariables #-}++module Main where+ import Sibe+ import Sibe.Word2Vec+ import Sibe.Utils+ import Data.Default.Class+ import qualified Data.Vector.Storable as V+ import Data.List (sortBy)+ import Data.Function (on)+ import Numeric.LinearAlgebra+ import System.IO+ import System.Directory+ import Data.List.Split+ import Control.Exception (evaluate)+ import Debug.Trace+ import Data.Char+ import System.Random++ rf :: FilePath -> IO String+ rf p = do+ hs <- openFile p ReadMode+ hSetEncoding hs latin1+ content <- evaluate =<< hGetContents hs+ length content `seq` hClose hs+ return content++ main = do+ setStdGen (mkStdGen 100)+ sws <- lines <$> readFile "examples/stopwords"++ -- real data, takes a lot of time to train+ ds <- do+ files <- filter ((/= "xml") . take 1 . reverse) <$> listDirectory "examples/blogs-corpus/"+ contents <- mapM (rf . ("examples/blogs-corpus/" ++)) files++ let texts = map (unwords . splitOn " ") contents+ let tags = ["<Blog>", "</Blog>", "<date>", "</date>", "<post>", "</post>", " "]+ return $ map cleanText $ removeWords (sws ++ tags) texts++ {-let ds = ["the king loves the queen", "the queen loves the king",-}+ {-"the dwarf hates the king", "the queen hates the dwarf",-}+ {-"the dwarf poisons the king", "the dwarf poisons the queen",-}+ {-"the man loves the woman", "the woman loves the man",-}+ {-"the thief hates the man", "the woman hates the thief",-}+ {-"the thief robs the man", "the thief robs the woman"]-}++ let session = def { learningRate = 5e-1+ , batchSize = 1+ , epochs = 200+ , debug = True+ } :: Session+ w2v = def { docs = ds+ , dimensions = 300+ , method = SkipGram+ , window = 2+ , w2vDrawChart = True+ , w2vChartName = "w2v-big-data.png"+ } :: Word2Vec++ (computed, vocvec) <- word2vec w2v session++ return ()++ cleanText :: String -> String+ cleanText string = + let notag = unwords $ filter ((/= "<date>") . take 6) (words string)+ ws = unwords $ filter (`notElem` ["urlLink"]) (words notag)+ spacify = foldl (\acc x -> replace x ' ' acc) (trim ws) [',', '/', '-', '\n', '\r', '?', '.', '(', ')', '%', '$', '"', ';', ':', '!', '\'']+ nonumber = filter (not . isNumber) spacify+ lower = map toLower nonumber+ in unwords . words $ lower+ where+ trim = f . f+ where+ f = reverse . dropWhile isSpace+ replace needle replacement =+ map (\c -> if c == needle then replacement else c)++ removeWords :: [String] -> [String] -> [String]+ removeWords ws documents =+ map rm documents+ where+ rm text = + unwords $ filter (`notElem` ws) (words text)+
+ examples/xor.hs view
@@ -0,0 +1,40 @@+module Main where+ import Sibe+ import Numeric.LinearAlgebra+ import Data.List+ import Debug.Trace+ import Data.Default.Class++ main = do+ let a = (sigmoid, sigmoid')+ rnetwork = randomNetwork 0 (-1, 1) 2 [(2, a)] (1, a) -- two inputs, 8 nodes in a single hidden layer, 1 output++ inputs = [vector [0, 1], vector [1, 0], vector [1, 1], vector [0, 0]]+ labels = [vector [1], vector [1], vector [0], vector [0]]++ session = def { network = rnetwork+ , learningRate = 0.8+ , epochs = 1000+ , training = zip inputs labels+ , test = zip inputs labels+ } :: Session++ initialCost = crossEntropy session++ newsession <- run gd session++ let results = map (`forward` newsession) inputs+ rounded = map (map round . toList) results++ cost = crossEntropy newsession++ putStrLn "parameters: "+ putStrLn $ "- inputs: " ++ show inputs+ putStrLn $ "- labels: " ++ show labels+ putStrLn $ "- learning rate: " ++ show (learningRate session)+ putStrLn $ "- epochs: " ++ show (epochs session)+ putStrLn $ "- initial cost (cross-entropy): " ++ show initialCost+ putStrLn "results: "+ putStrLn $ "- actual result: " ++ show results+ putStrLn $ "- rounded result: " ++ show rounded+ putStrLn $ "- cost (cross-entropy): " ++ show cost
+ sibe.cabal view
@@ -0,0 +1,111 @@+name: sibe+version: 0.1.0.0+synopsis: Initial project template from stack+description: Haskell Machine Learning+homepage: https://github.com/mdibaiee/sibe+license: GPL-3+license-file: LICENSE+author: Mahdi Dibaiee+maintainer: mdibaiee@aol.com+copyright: 2016 Mahdi Dibaiee+category: Web, Machine Learning, Data Science+build-type: Simple+-- extra-source-files:+cabal-version: >=1.10++library+ hs-source-dirs: src+ exposed-modules: Sibe, Sibe.NaiveBayes, Sibe.NLP, Sibe.Word2Vec, Sibe.Utils+ build-depends: base >= 4.7 && < 5+ , hmatrix+ , random+ , deepseq+ , containers+ , split+ , regex-base+ , regex-pcre+ , text+ , stemmer+ , vector+ , random-shuffle+ , data-default-class+ , Chart+ , Chart-cairo+ , lens+ default-language: Haskell2010++executable example-xor+ hs-source-dirs: examples+ main-is: xor.hs+ ghc-options: -threaded -rtsopts -with-rtsopts=-N+ build-depends: base+ , sibe+ , hmatrix+ , data-default-class+ default-language: Haskell2010++executable example-word2vec+ hs-source-dirs: examples+ main-is: word2vec.hs+ ghc-options: -threaded -rtsopts -with-rtsopts=-N+ build-depends: base+ , sibe+ , hmatrix+ , data-default-class+ , split+ , vector+ , directory+ , random+ default-language: Haskell2010++executable example-424+ hs-source-dirs: examples+ main-is: 424encoder.hs+ ghc-options: -threaded -rtsopts -with-rtsopts=-N+ build-depends: base+ , sibe+ , hmatrix+ , data-default-class+ default-language: Haskell2010++executable example-notmnist+ hs-source-dirs: examples+ main-is: notmnist.hs+ ghc-options: -threaded -rtsopts -with-rtsopts=-N+ build-depends: base+ , sibe+ , hmatrix+ , directory >= 1.2.5.0+ , JuicyPixels == 3.2.7.2+ , vector == 0.11.0.0+ , random+ , random-shuffle+ , data-default-class+ , Chart+ , Chart-cairo+ default-language: Haskell2010++executable example-naivebayes-doc-classifier+ hs-source-dirs: examples+ main-is: naivebayes-doc-classifier.hs+ ghc-options: -threaded -rtsopts -with-rtsopts=-N+ build-depends: base+ , sibe+ , hmatrix+ , containers+ , split+ default-language: Haskell2010++test-suite sibe-test+ type: exitcode-stdio-1.0+ hs-source-dirs: test+ main-is: Spec.hs+ build-depends: base+ , sibe+ , hmatrix+ ghc-options: -threaded -rtsopts -with-rtsopts=-N+ default-language: Haskell2010++source-repository head+ type: git+ location: https://github.com/mdibaiee/sibe
+ src/Sibe.hs view
@@ -0,0 +1,362 @@+{-# LANGUAGE GADTs #-}+{-# LANGUAGE BangPatterns #-}+{-# LANGUAGE DataKinds #-}+{-# LANGUAGE TypeOperators #-}++module Sibe+ (Network(..),+ Layer(..),+ Input,+ Output,+ Activation,+ forward,+ forward',+ runLayer,+ runLayer',+ randomLayer,+ randomNetwork,+ buildNetwork,+ saveNetwork,+ loadNetwork,+ train,+ gd,+ sgd,+ run,+ sigmoid,+ sigmoid',+ softmax,+ softmax',+ relu,+ relu',+ crossEntropy,+ crossEntropy',+ genSeed,+ replaceVector,+ Session(..),+ accuracy,+ learningRateDecay,+ ignoreBiases,+ one+ ) where+ import Numeric.LinearAlgebra+ import System.Random+ import System.Random.Shuffle+ import Debug.Trace+ import Data.List (foldl', sortBy, genericLength, permutations)+ import System.IO+ import Control.DeepSeq+ import Control.Monad+ import qualified Data.Vector.Storable as V+ import Data.Default.Class+ import System.Exit++ import qualified Graphics.Rendering.Chart.Easy as Chart+ import Graphics.Rendering.Chart.Backend.Cairo++ type LearningRate = Double+ type Input = Vector Double+ type Output = Vector Double+ type Activation = (Vector Double -> Vector Double, Vector Double -> Vector Double)++ data Layer = Layer { biases :: !(Vector Double)+ , nodes :: !(Matrix Double)+ , activation :: Activation+ }++ instance Show Layer where+ show (Layer biases nodes _) = "(" ++ show biases ++ "," ++ show nodes ++ ")"++ data Network = O Layer+ | Layer :- Network+ + instance Show Network where+ show (Layer biases nodes _ :- n) =+ (show . length $ toLists nodes) ++ "x" ++ (show . length . head . toLists $ nodes) ++ " " ++ (show . length . toList $ biases) ++ " :- " ++ show n+ show (O (Layer biases nodes _)) =+ (show . length $ toLists nodes) ++ "x" ++ (show . length . head . toLists $ nodes) ++ " " ++ (show . length . toList $ biases)+ + infixr 5 :-++ data Session = Session { network :: Network+ , training :: [(Vector Double, Vector Double)]+ , test :: [(Vector Double, Vector Double)]+ , learningRate :: Double+ , epochs :: Int+ , epoch :: Int+ , batchSize :: Int+ , chart :: [(Int, Double, Double)]+ , drawChart :: Bool+ , chartName :: String+ , momentum :: Double+ , debug :: Bool+ } deriving (Show)++ emptyNetwork = randomNetwork 0 (0, 0) 0 [] (0, (id, id))+ instance Default Session where+ def = Session { network = seq (die "You have not specified a network parameter") emptyNetwork+ , training = seq (die "You have not specified training data") []+ , test = seq (die "You have not specified test data") []+ , learningRate = 0.5+ , epochs = 35+ , epoch = 0+ , batchSize = 0+ , chart = []+ , drawChart = False+ , chartName = "chart.png"+ , momentum = 0+ , debug = False+ }++ saveNetwork :: Network -> String -> IO ()+ saveNetwork network file =+ writeFile file ((show . reverse) (gen network []))+ where+ gen (O (Layer biases nodes _)) list = (biases, nodes) : list+ gen (Layer biases nodes _ :- n) list = gen n $ (biases, nodes) : list++ loadNetwork :: [Activation] -> String -> IO Network+ loadNetwork activations file = do+ handle <- openFile file ReadMode+ content <- hGetContents handle+ let list = read content :: [(Vector Double, Matrix Double)]+ network = gen list activations+ content `deepseq` hClose handle+ return network++ where+ gen [(biases, nodes)] [a] = O (Layer biases nodes a)+ gen ((biases, nodes):hs) (a:as) = Layer biases nodes a :- gen hs as++ runLayer :: Input -> Layer -> Output+ runLayer input (Layer !biases !weights _) = input <# weights + biases++ runLayer' :: Input -> Layer -> Output+ runLayer' input (Layer !biases !weights _) = input <# weights++ forward :: Input -> Session -> Output+ forward input session = compute input (network session)+ where+ compute input (O l@(Layer _ _ (fn, _))) = fn $ runLayer input l+ compute input (l@(Layer _ _ (fn, _)) :- n) = compute ((fst . activation $ l) $ runLayer input l) n++ forward' :: Input -> Session -> Output+ forward' input session = compute input (network session)+ where+ compute input (O l@(Layer _ _ (fn, _))) = fn $ runLayer' input l+ compute input (l@(Layer _ _ (fn, _)) :- n) = compute ((fst . activation $ l) $ runLayer' input l) n++ randomLayer :: Seed -> (Int, Int) -> (Double, Double) -> Activation -> Layer+ randomLayer seed (wr, wc) (l, u) =+ let weights = uniformSample seed wr $ replicate wc (l, u)+ biases = randomVector seed Uniform wc * realToFrac u - realToFrac l+ in Layer biases weights++ randomNetwork :: Seed -> (Double, Double) -> Int -> [(Int, Activation)] -> (Int, Activation) -> Network+ randomNetwork seed bound input [] (output, a) =+ O $ randomLayer seed (input, output) bound a+ randomNetwork seed bound input ((h, a):hs) output =+ randomLayer seed (input, h) bound a :-+ randomNetwork (seed + 1) bound h hs output++ buildNetwork :: Seed -> (Double, Double) -> Int -> [(Int, Int, Activation)] -> (Int, Int, Activation) -> Network+ buildNetwork seed bound input [] (outputRows, outputColumns, a) =+ O $ randomLayer seed (input, outputColumns) bound a+ buildNetwork seed bound input ((rows, columns, a):hs) output =+ randomLayer seed (input, columns) bound a :-+ buildNetwork (seed + 1) bound columns hs output++ sigmoid :: Vector Double -> Vector Double+ sigmoid x = 1 / max (1 + exp (-x)) 1e-10++ sigmoid' :: Vector Double -> Vector Double+ sigmoid' x = sigmoid x * (1 - sigmoid x)++ softmax :: Vector Double -> Vector Double+ softmax x = cmap (\a -> exp a / s) x+ where+ s = V.sum $ exp x++ softmax' :: Vector Double -> Vector Double+ softmax' = cmap (\a -> sig a * (1 - sig a))+ where+ sig x = 1 / max (1 + exp (-x)) 1e-10++ -- used for negative sampling+ {-sampledSoftmax :: Vector Double -> Vector Double-}+ {-sampledSoftmax x = cmap (\a -> exp a / s) x-}+ {-where-}+ {-s = V.sum . exp $ x-}++ relu :: Vector Double -> Vector Double+ relu = cmap (max 0.1)++ relu' :: Vector Double -> Vector Double+ relu' = cmap dev+ where dev x+ | x < 0 = 0+ | otherwise = 1++ crossEntropy :: Session -> Double+ crossEntropy session =+ let inputs = map fst (test session)+ labels = map (toList . snd) (test session)+ outputs = map (toList . (`forward` session)) inputs+ pairs = zip outputs labels+ n = genericLength pairs+ in sum (map set pairs) / n+ where+ set (os, ls) = (-1 / genericLength os) * sum (zipWith f os ls)+ f a y = y * log (max 1e-10 a)++ crossEntropy' :: Vector Double -> Vector Double+ crossEntropy' x = 1 / fromIntegral (V.length x)++ one :: Vector Double -> Vector Double+ one v = vector $ replicate (V.length v) 1++ train :: Input+ -> Network+ -> Output -- target+ -> Double -- learning rate+ -> Network -- network's output+ train input network target alpha = fst $ run input network+ where+ run :: Input -> Network -> (Network, Vector Double)+ run input (O l@(Layer biases weights (fn, fn'))) =+ let y = runLayer input l+ o = fn y+ delta = o - target + de = delta * fn' y++ biases' = biases - scale alpha de+ weights' = weights - scale alpha (input `outer` de) -- small inputs learn slowly+ layer = Layer biases' weights' (fn, fn') -- updated layer++ pass = weights #> de++ in (O layer, pass)+ run input (l@(Layer biases weights (fn, fn')) :- n) =+ let y = runLayer input l+ o = fn y+ (n', delta) = run o n++ de = delta * fn' y++ biases' = biases - scale alpha de+ weights' = weights - scale alpha (input `outer` de)+ layer = Layer biases' weights' (fn, fn')++ pass = weights #> de+ in (layer :- n', pass)++ gd :: Session -> IO Session+ gd session = do+ seed <- newStdGen++ let pairs = training session+ alpha = learningRate session+ net = network session++ let n = length pairs++ shuffled <- shuffleM pairs++ let newnet = foldl' (\n (input, label) -> train input n label alpha) net pairs+ cost = crossEntropy (session { network = newnet })++ let el = map (\(e, l, _) -> (e, l)) (chart session)+ ea = map (\(e, _, a) -> (e, a)) (chart session)++ when (drawChart session) $ do+ toFile Chart.def (chartName session) $ do+ Chart.layoutlr_title Chart..= "loss over time"+ Chart.plotLeft (Chart.line "loss" [el])+ Chart.plotRight (Chart.line "learningRate" [ea])++ return session { network = newnet+ , epoch = epoch session + 1+ , chart = (epoch session, cost, learningRate session):chart session+ }++ sgd :: Session -> IO Session+ sgd session = do+ seed <- newStdGen++ let pairs = training session+ bsize = batchSize session+ alpha = learningRate session+ net = network session++ let n = length pairs+ iterations = n `div` bsize - 1++ shuffled <- shuffleM pairs++ let iter net i =+ let n = length pairs+ batch = take bsize . drop (i * bsize) $ shuffled+ batchInputs = map fst batch+ batchLabels = map snd batch+ batchPair = zip batchInputs batchLabels+ in foldl' (\n (input, label) -> train input n label alpha) net batchPair++ let newnet = foldl' iter net [0..iterations]+ cost = crossEntropy (session { network = newnet })++ let el = map (\(e, l, _) -> (e, l)) (chart session)+ ea = map (\(e, _, a) -> (e, a)) (chart session)++ when (drawChart session) $ do+ toFile Chart.def (chartName session) $ do+ Chart.layoutlr_title Chart..= "loss over time"+ Chart.plotLeft (Chart.line "loss" [el])+ Chart.plotRight (Chart.line "learningRate" [ea])++ return session { network = newnet+ , epoch = epoch session + 1+ , chart = (epoch session, cost, learningRate session):chart session+ }+++ accuracy :: Session -> Double+ accuracy session = + let inputs = map fst (test session)+ labels = map snd (test session)++ results = map (`forward` session) inputs+ rounded = map (map round . toList) results++ equals = zipWith (==) rounded (map (map round . toList) labels)+ in genericLength (filter (== True) equals) / genericLength inputs++ learningRateDecay :: (Double, Double) -> Session -> Session+ learningRateDecay (step, m) session =+ session { learningRate = max m $ learningRate session / step }++ ignoreBiases :: Session -> Session+ ignoreBiases session =+ session { network = rmbias (network session) }+ where+ rmbias (O (Layer biases nodes a)) = O $ Layer (biases * 0) nodes a+ rmbias ((Layer biases nodes a) :- n) = Layer (biases * 0) nodes a :- rmbias n++ run :: (Session -> IO Session)+ -> Session -> IO Session+ run fn session = foldM (\s i -> fn s) session [0..epochs session]++ factorial :: Int -> Int+ factorial 0 = 1+ factorial x = x * factorial (x - 1)++ genSeed :: IO Seed+ genSeed = do+ (seed, _) <- random <$> newStdGen :: IO (Int, StdGen)+ return seed++ replaceVector :: Vector Double -> Int -> Double -> Vector Double+ replaceVector vec index value =+ let list = toList vec+ in fromList $ take index list ++ value : drop (index + 1) list++ clip :: Double -> (Double, Double) -> Double+ clip x (l, u) = min u (max l x)
+ src/Sibe/NLP.hs view
@@ -0,0 +1,122 @@+module Sibe.NLP+ (Class,+ Document(..),+ accuracy,+ recall,+ precision,+ fmeasure,+ cleanText,+ cleanDocuments,+ removeWords,+ removeStopwords,+ ngram,+ ngramText,+ )+ where+ import Sibe.Utils+ import Data.List+ import Debug.Trace+ import Data.List.Split+ import Data.Maybe+ import Control.Arrow ((&&&))+ import Text.Regex.PCRE+ import Data.Char (isSpace, isNumber, toLower)+ import NLP.Stemmer+ import qualified Data.Set as Set++ type Class = Int;++ data Document = Document { text :: String+ , c :: Class+ } deriving (Eq, Show, Read)+++ cleanText :: String -> String+ cleanText string =+ let puncs = filter (`notElem` ['!', '"', '#', '$', '%', '(', ')', '.', '?']) (trim string)+ spacify = foldl (\acc x -> replace x ' ' acc) puncs [',', '/', '-', '\n', '\r']+ stemmed = unwords $ map (stem Porter) (words spacify)+ nonumber = filter (not . isNumber) stemmed+ lower = map toLower nonumber+ in (unwords . words) lower -- remove unnecessary spaces+ where+ trim = f . f+ where+ f = reverse . dropWhile isSpace+ replace needle replacement =+ map (\c -> if c == needle then replacement else c)++ cleanDocuments :: [Document] -> [Document]+ cleanDocuments documents =+ let cleaned = map (\(Document text c) -> Document (cleanText text) c) documents+ in cleaned++ removeWords :: [String] -> [Document] -> [Document]+ removeWords ws documents =+ map (\(Document text c) -> Document (rm ws text) c) documents+ where+ rm list text =+ unwords $ filter (`notElem` list) (words text)++ removeStopwords :: Int -> [Document] -> [Document]+ removeStopwords i documents =+ let wc = wordCounts (concatDocs documents)+ wlist = sortBy (\(_, a) (_, b) -> b `compare` a) wc+ stopwords = map fst (take i wlist)+ in removeWords stopwords documents+ where+ vocabulary x = ordNub (words x)+ countWordInDoc d w = genericLength (filter (==w) d)+ wordCounts x =+ let voc = vocabulary x+ in zip voc $ map (countWordInDoc (words x)) voc++ concatDocs = concatMap (\(Document text _) -> text ++ " ")++ accuracy :: [(Int, (Int, Double))] -> Double+ accuracy results =+ let pairs = map (\(a, b) -> (a, fst b)) results+ correct = filter (uncurry (==)) pairs+ in genericLength correct / genericLength results++ recall :: [(Int, (Int, Double))] -> Double+ recall results =+ let classes = ordNub (map fst results)+ s = sum (map rec classes) / genericLength classes+ in s+ where+ rec a =+ let t = genericLength $ filter (\(c, (r, _)) -> c == r && c == a) results+ y = genericLength $ filter (\(c, (r, _)) -> c == a) results+ in t / y++ precision :: [(Int, (Int, Double))] -> Double+ precision results =+ let classes = ordNub (map fst results)+ s = sum (map prec classes) / genericLength classes+ in s+ where+ prec a =+ let t = genericLength $ filter (\(c, (r, _)) -> c == r && c == a) results+ y = genericLength $ filter (\(c, (r, _)) -> r == a) results+ in+ if y == 0+ then 0+ else t / y++ fmeasure :: [(Int, (Int, Double))] -> Double+ fmeasure results =+ let r = recall results+ p = precision results+ in (2 * p * r) / (p + r)++ ngram :: Int -> [Document] -> [Document]+ ngram n documents =+ map (\(Document text c) -> Document (ngramText n text) c) documents++ ngramText :: Int -> String -> String+ ngramText n text =+ let ws = words text+ pairs = zip [0..] ws+ grams = map (\(i, w) -> concat . intersperse "_" $ w:((take (n - 1) . drop (i+1)) ws)) pairs+ in unwords ("<b>_":grams)
+ src/Sibe/NaiveBayes.hs view
@@ -0,0 +1,128 @@+module Sibe.NaiveBayes+ (Document(..),+ NB(..),+ initialize,+ run,+ session,+ accuracy,+ precision,+ recall,+ fmeasure,+ mean,+ stdev,+ cleanText,+ cleanDocuments,+ ngram,+ ngramText,+ removeWords,+ removeStopwords,+ )+ where+ import Sibe.Utils+ import Sibe.NLP+ import Data.List+ import Debug.Trace+ import qualified Data.Set as Set+ import Data.List.Split+ import Data.Maybe+ import Control.Arrow ((&&&))++ data NB = NB { documents :: [Document]+ , classes :: [(Class, Double)]+ , vocabulary :: Int+ , megadoc :: String+ , cd :: [(Class, [Document])]+ , cw :: [(Class, [(String, Int)])]+ , cgram :: [(Class, [(String, Int)])]+ } deriving (Eq, Show, Read)++ initialize :: [Document] -> [Class] -> NB+ initialize documents classes =+ let megadoc = concatDocs documents+ vocabulary = genericLength ((ordNub . words) megadoc)+ -- (class, prior probability)+ cls = zip classes (map classPrior classes)++ -- (class, [document])+ cd = zip classes (map classDocs classes)++ -- (class, [(word, count)])+ cw = zip classes $ map classWordsCounts classes++ cgram = zip classes $ map classNGramCounts classes++ in NB { documents = documents+ , classes = cls+ , vocabulary = vocabulary+ , megadoc = megadoc+ , cd = cd+ , cw = cw+ , cgram = cgram+ }+ where+ concatDocs = concatMap (\(Document text _) -> text ++ " ")++ classDocs x = filter ((==x) . c) documents+ classMegadoc = concatMap (\(Document text _) -> text ++ " ") . classDocs+ classWords = words . classMegadoc+ classNGram = concatMap (\(Document text _) -> text ++ " ") . ngram 2 . classDocs+ classNGramWords = words . classNGram+ classVocabulary = ordNub . classWords+ classPrior x = genericLength (classDocs x) / genericLength documents+ countWordInDoc d w = genericLength (filter (==w) d)+ wordsCount ws voc =+ zip voc $ map (countWordInDoc ws) voc+ classWordsCounts x = wordsCount (classWords x) (classVocabulary x)+ classNGramCounts x = wordsCount (classNGramWords x) (ordNub $ classNGramWords x)++ session :: [Document] -> NB -> [(Class, (Class, Double))]+ session docs nb =+ let results = map (\(Document text c) -> (c, run text nb)) docs+ in results++ run :: String -> NB -> (Class, Double)+ run txt (NB documents classes vocabulary megadoc cd cw cgram) =+ let scores = map (score . fst) classes+ index = argmax scores+ m = maximum scores+ in (fst (classes !! index), m)+ where+ score c =+ let prior = snd (fromJust $ find ((==c) . fst) classes)++ -- below is the formula according to Multinominal Naive Bayes, but it seems+ -- using a uniform prior probability seems to work better when working with imbalanced+ -- training datasets, instead, we help rare classes get higher scores using+ -- alpha = (1 - prior * ALPHA), we use ALPHA = 1 here+ -- in prior * product (map (prob c) (words txt))++ alpha = 1 - prior++ in alpha * product (map (prob c) (words txt))++ prob c w =+ let fcw = fromJust $ find ((==c) . fst) cw+ fcg = fromJust $ find ((==c) . fst) cgram+ tctM = find ((== w) . fst) (snd fcw)+ tct = if isJust tctM then (snd . fromJust) tctM else 0+ cvoc = sum $ map snd (snd fcw)+ voc = vocabulary+ gram = find ((==w) . last . splitOn "_" . fst) (snd fcg)+ pg = if isJust gram then (snd . fromJust) gram else 0+ -- in realToFrac (tct * pg + 1) / realToFrac (cvoc + voc) -- uncomment to enable ngrams+ in realToFrac (tct + 1) / realToFrac (cvoc + voc)++ argmax :: (Ord a) => [a] -> Int+ argmax x = fst $ maximumBy (\(_, a) (_, b) -> a `compare` b) (zip [0..] x)++ mean :: [Double] -> Double+ mean x = sum x / genericLength x++ stdev :: [Double] -> Double+ stdev x =+ let avg = mean x+ variance = sum (map ((^2) . subtract avg) x) / (genericLength x - 1)+ in sqrt variance++ l :: (Show a) => a -> a+ l a = trace (show a) a
+ src/Sibe/Utils.hs view
@@ -0,0 +1,28 @@+module Sibe.Utils+ ( similarity+ , ordNub+ , onehot+ , average+ ) where+ import qualified Data.Vector.Storable as V+ import qualified Data.Set as Set+ import Numeric.LinearAlgebra++ similarity :: Vector Double -> Vector Double -> Double+ similarity a b = (V.sum $ a * b) / (magnitude a * magnitude b)+ where+ magnitude :: Vector Double -> Double+ magnitude v = sqrt $ V.sum (cmap (^2) v)++ onehot :: Int -> Int -> Vector Double+ onehot len i = vector $ replicate i 0 ++ [1] ++ replicate (len - i - 1) 0++ ordNub :: (Ord a) => [a] -> [a]+ ordNub = go Set.empty+ where+ go _ [] = []+ go s (x:xs) = if x `Set.member` s then go s xs+ else x : go (Set.insert x s) xs++ average :: Vector Double -> Vector Double+ average v = cmap (/ (V.sum v)) v
+ src/Sibe/Word2Vec.hs view
@@ -0,0 +1,131 @@+module Sibe.Word2Vec+ ( word2vec+ , Word2Vec (..)+ , W2VMethod (..)+ ) where+ import Sibe+ import Sibe.Utils+ import Debug.Trace+ import Data.Char+ import Data.Maybe+ import Data.List+ import Numeric.LinearAlgebra hiding (find)+ import qualified Data.Vector.Storable as V+ import Data.Default.Class+ import Data.Function (on)+ import Control.Monad+ import System.Random++ import Graphics.Rendering.Chart as Chart+ import Graphics.Rendering.Chart.Backend.Cairo+ import Control.Lens++ data W2VMethod = SkipGram | CBOW+ data Word2Vec = Word2Vec { docs :: [String]+ , window :: Int+ , dimensions :: Int+ , method :: W2VMethod+ , w2vChartName :: String+ , w2vDrawChart :: Bool+ }+ instance Default Word2Vec where+ def = Word2Vec { docs = []+ , window = 2+ , w2vChartName = "w2v.png"+ , w2vDrawChart = False+ }++ word2vec w2v session = do+ seed <- newStdGen++ let s = session { training = trainingData+ , network = randomNetwork 0 (-1, 1) v [(dimensions w2v, (id, one))] (v, (softmax, crossEntropy'))+ }++ when (debug s) $ do+ putStr "vocabulary size: "+ print v++ putStr "trainingData length: "+ print . length $ trainingData++ -- biases are not used in skipgram/cbow+ newses <- run (sgd . ignoreBiases) s+++ -- export the hidden layer+ let (hidden@(Layer biases nodes _) :- _) = network newses+ -- run words through the hidden layer alone to get the word vector+ let computedVocVec = map (\(w, v) -> (w, runLayer' v hidden)) vocvec++ when (w2vDrawChart w2v) $ do+ let mat = fromColumns . map snd $ computedVocVec+ (u, s, v) = svd mat+ cut = subMatrix (0, 0) (2, cols mat)+ diagS = diagRect 0 (V.take 2 s) (rows mat) (cols mat)++ twoDimensions = cut $ u <> diagS <> tr v+ textData = zipWith (\s l -> (V.head l, V.last l, s)) (map fst computedVocVec) (toColumns twoDimensions)++ chart = toRenderable layout+ where+ textP = plot_annotation_values .~ textData+ $ def+ layout = layout_title .~ "word vectors"+ $ layout_plots .~ [toPlot textP]+ $ def+ + renderableToFile def (w2vChartName w2v) chart+ return ()++ return (computedVocVec, vocvec)+ where+ -- clean documents+ ds = map cleanText (docs w2v)++ -- words of each document+ wd = map (words . (++ " ") . (map toLower)) ds++ -- all words together, used to generate the vocabulary+ ws = words (concatMap ((++ " ") . map toLower) ds)+ vocabulary = ordNub ws+ v = length vocabulary++ -- generate one-hot vectors for each word of vocabulary+ vocvec = zip vocabulary $ map (onehot v) [0..v - 1]++ -- training data: generate input and output pairs for each word and the words in it's window+ trainingData = concatMap (\wds -> concatMap (iter wds) $ zip [0..] wds) wd+ where+ iter wds (i, w) =+ let v = snd . fromJust . find ((==w) . fst) $ vocvec+ before = take (window w2v) . drop (i - window w2v) $ wds+ after = take (window w2v) . drop (i + 1) $ wds+ ns + | i == 0 = after+ | i == length vocvec - 1 = before+ | otherwise = before ++ after+ vectorized = map (\w -> snd . fromJust $ find ((== w) . fst) vocvec) ns+ new = foldl1 (+) vectorized+ in+ if length wds <= 1+ then []+ else+ case method w2v of+ SkipGram -> [(v, average new)]+ CBOW -> [(average new, v)]+ _ -> error "unsupported word2vec method"++ cleanText :: String -> String+ cleanText string =+ let puncs = filter (`notElem` ['!', '"', '#', '$', '%', '(', ')', '.', '?', '\'']) (trim string)+ spacify = foldl (\acc x -> replace x ' ' acc) puncs [',', '/', '-', '\n', '\r']+ nonumber = filter (not . isNumber) spacify+ lower = map toLower nonumber+ in (unwords . words) lower -- remove unnecessary spaces+ where+ trim = f . f+ where+ f = reverse . dropWhile isSpace+ replace needle replacement =+ map (\c -> if c == needle then replacement else c)
+ test/Spec.hs view
@@ -0,0 +1,8 @@+module Main where+ import System.Exit (exitFailure)++ import Sibe++ main = do+ putStrLn "Hey"+ exitFailure