…ature branch (#3324) * Initial commit * ci test build * forgot to save this one file * Debug-Intrinsics isn't a valid config, trying windows-x64 * disabled tests for now * disable tests attempt 2 * initial code push, no history, test project not in the build so is the internal client * battling with warn as err * test build * test change * make params for MLContext data extensions match ML.NET default names and values; update gitignore; nit rev for Benchmarking.cs (#5) * Create README.md (#2) * API folder changes (#6) * comment out fast forest trainer, per discussion on ML.NET open issue #1983, for now, to run E2E w/o exceptions (#7) * Make validation data param mandatory; remove GetFirstPipeline sample (#10) * Make validation data param mandatory; remove GetFirstPipeline sample * remove deprecated todo * Create ISSUE_TEMPLATE.md & PULL_REQUEST_TEMPLATE.md (#12) * Create ISSUE_TEMPLATE.md * Create PULL_REQUEST_TEMPLATE.md * NestedObject For pipeline (#14) * add estimator extensions / catalog; add conversion from external to internal pipeline; transform clean-up; add back in test proj and fix build; refactor trainer ext name mappings (#15) * Make validation data param mandatory; remove GetFirstPipeline sample * remove deprecated todo * add estimator extensions / catalog; add ability to go from external to internal pipeline; a lot of transform clean-up; add back in test proj and get it building; refactor trainer ext name mappings * corrected the typo in readme (#16) * make GetNextPipeline API w/ public Pipeline method on PipelineSuggester; write GetNextPipeline API test; fix public Pipeline object serialization; fix header inferencing bug; write test utils for fetching datasets (#18) * get next pipeline API rev -- refactor API to consume column dimensions, purpose, type, and name instead of available trainers & transforms (#19) * mark get next pipeline test as ignore for now (#20) * fix dataview take util bug, add dataview skip util, add some UTs to increase code coverage (#21) * fix dataview take util bug, add dataview skip util, add some UTs to increase code coverage * add accuracy threshold on AutoFit test * add null check to best pipeline on autofit result * unit test additions (including user input validation testing); dead code removal for code coverage (including KDO & associated utils); misc fixes & revs (#22) * add trainer extension tests, & misc fixes (#23) * add estimator extension tests (#24) * add conversions tests (#25) * fix multiclass runs & add multiclass autofit UT (#27) * add basic autofit regression test (#28) * fix categorical transform bug (sometimes categorical features weren't concatenated to final features); add UT transforms; add PipelineNode equality & tests to serve as AutoML testing infra * add example to readme (#26) * add lightgbm args as nested properties (#33) * fix bug where if one pipeline hyperparam optimization converges, run terminates (#36) * add open-source headers to files; other nit clean-ups along the way (#35) * Ungroup Columns in Column Inference (#40) * Added sequential grouping of columns * added ungrouping of column option * reverted the file * Misc fixes (#39) * misc fixes -- fix bug where SMAC returning already-seen values; fix param encoding return bug in pipeline object model; nit clean-up AutoFit; return in pipeline suggester when sweeper has no next proposal; null ref fix in public object model pipeline suggester * fix in BuildPipelineNodePropsLightGbm test, fix / use correct 'newTrainer' variable in PipelneSuggester * SMAC perf improvement * Removing the nuget.config and have build.props mention the nuget package sources. (#38) * Added sequential grouping of columns * removed nuget.config and have only props mentions the nuget sources * reverted the file * transform inferencing concat / ignore fixes (#41) * make pipeline object model & other public classes internal (#43) * handle SMAC exception when fewer trees were trained than requested (#44) * Throw error on incorrect Label name in InferColumns API (#47) * Added sequential grouping of columns * reverted the file * addded infer columns label name checking * added column detection error * removed unsed usings * added quotes * replace Where with Any clause * replace Where with Any clause * Set Nullable Auto params to null values (#50) * Added sequential grouping of columns * reverted the file * added auto params as null * change to the update fields method * First public api propsal (#52) * Includes following 1) Final proposal for 0.1 public API surface 2) Prefeaturization 3) Splitting train data into train and validate when validation data is null 4) Providing end to end samples one each for regression, binaryclassification and multiclass classification * Incorporating code review feedbacks * Revert "Set Nullable Auto params to null values" (#53) * Revert "First public api propsal (#52)" This reverts commit e4a64cf. * Revert "Set Nullable Auto params to null values (#50)" This reverts commit 41c663c. * AutoFit return type is now an IEnumerable (#55) AutoFit returns is now an IEnumerable - this enables many good things Implementing variety of early stopping criteria (See sample) Early discard of models that are no good. This improves memory usage efficiency. (See sample) No need to implement a callback to get results back Getting best score is now outside of API implementation. It is a simple math function to compare scores (See sample). Also templatized the return type for better type safety through out the code. * misc fixes & test additions, towards 0.1 release (#56) * Enable UnitTests on build server (#57) * 1) Making trainer name public (#62) 2) Fixing up samples to reflect it * Initial version of CLI tool for mlnet (#61) * added global tool initial project * removed unneccesary files, renamed files * refactoring and added base abstract classes for trainer generator * removed unused class * Added classes for transforms * added transform generate dummy classes * more refactoring, added first transform * more refactoring and added classes * changed the project structure * restructing added options class * sln changes * refactored options to different class: * added more logic for code generation of class * misc changes * reverted file * added commandline api package * reverted sample * added new command line api parser * added normalization of column names * Added command defaults and error message * implementation of all trainers * changed auto to null * added all transform generators * added error handling when args is empty and minor changes due to change in AutoML api names * changed the name of param * added new command line options and restructuring code * renamed proj file and added solution * Added code to generate usings, Fixed few bugs in the code * added validation to the command line options * changed project name * Bug fixes due to API change in AutoML * changed directory structure * added test framework and basic tests * added more tests * added improvements to template and error handling * renamed the estimator name * fixed test case * added comments * added headers * changed namespace and removed unneccesary properties from project * Revert "changed namespace and removed unneccesary properties from project" This reverts commit 9edae033e9845e910f663f296e168f1182b84f5f. * fixed test cases and renamed namespaces * cleaned up proj file * added folder structure * added symbols/tokens for strings * added more tests * review comments * modified test cases * review comments * change in the exception message * normalized line endings * made method private static * simplified range building /optimization * minor fix * added header * added static methods in command where necessary * nit picks * made few methods static * review comments * nitpick * remove line pragmas * fix test case * Use better AutiFit overload and ignore Multiclass (#64) * Upgrading CLI to produce ML.NET V.10 APIs and bunch of Refactoring tasks (#65) * Added sequential grouping of columns * reverted the file * upgrade to v .10 and refactoring * added null check * fixed unit tests * review comments * removed the settings change * added regions * fixed unit tests * Upgrade ML.NET package to 0.10.0 (#70) * Change in template to accomodate new API of TextLoader (#72) * Added sequential grouping of columns * reverted the file * changed to new API of Text Loader * changed signature * added params for taking additional settings * changes to codegen params * refactoring of templates and fixing errors * Enable gated check for mlnet.tests (#79) * Added sequential grouping of columns * reverted the file * changed to new API of Text Loader * changed signature * added params for taking additional settings * changes to codegen params * refactoring of templates and fixing errors * added run-tests.proj and referred it in build.proj * CLI tool - make validation dataset optional and support for crossvalidation in generated code (#83) * Added sequential grouping of columns * reverted the file * bug fixes, more logic to templates to support cross-validate * formatting and fix type in consolehelper * Added logic in templates * revert settings * benchmarking related changes (#63) * Create test.txt * Create test.txt * changes needed for benchmarking * forgot one file * merge conflict fix * fix build break * back out my version of the fix for Label column issue and fix the original fix * bogus file removal * undo SuggestedPipeline change * remove labelCol from pipeline suggester * fix build break * fix fast forest learner (don't sweep over learning rate) (#88) * Made changes to Have non-calibrated scoring for binary classifiers (#86) * Added sequential grouping of columns * reverted the file * added calibration workaround * removed print probability * reverted settings * rev ColumnInference API: can take label index; rev output object types; add tests (#89) * rename AutoML to Microsoft.ML.Auto everywhere and a shot at publishing nuget package (#99) * Create test.txt * Create test.txt * changes needed for benchmarking * forgot one file * merge conflict fix * fix build break * back out my version of the fix for Label column issue and fix the original fix * bogus file removal * undo SuggestedPipeline change * remove labelCol from pipeline suggester * fix build break * rename AutoML to Microsoft.ML.Auto everywhere and a shot at publishing nuget package (will probably need tweaks once I try to use the pipleline) * publish nuget (#101) * use dotnet-internal-temp agent for internal build * use dotnet-internal feed * Fix Codegen for columnConvert and ValueToKeyMapping transform and add individual transform tests (#95) * Added sequential grouping of columns * reverted the file * fix usings for type convert * added transforms tests * review comments * When generating usings choose only distinct usings directives (#94) * Added sequential grouping of columns * reverted the file * Added code to have unique strings * refactoring * minor fix * minor fix * Autofit overloads + cancellation + progress callbacks 1) Introduce AutoFit overloads (basic and advanced) 2) AutoFit Cancellation 3) AutoFit progress callbacks * Default the kfolds to value 5 in CLI generated code (#115) * Added sequential grouping of columns * reverted the file * Set up CI with Azure Pipelines * Update azure-pipelines.yml for Azure Pipelines * Update azure-pipelines.yml for Azure Pipelines * remove file * added kfold param and defaulted to value * changed type * added for regression * Remove extra ; from generated code (#114) * Added sequential grouping of columns * reverted the file * Set up CI with Azure Pipelines * Update azure-pipelines.yml for Azure Pipelines * Update azure-pipelines.yml for Azure Pipelines * removed extra ; from generated code * removed file * fix unit tests * TimeoutInSeconds (#116) Specifying timeout in seconds instead of minutes * Added more command line args implementation to CLI tool and refactoring (#110) * Added sequential grouping of columns * reverted the file * Set up CI with Azure Pipelines * Update azure-pipelines.yml for Azure Pipelines * Update azure-pipelines.yml for Azure Pipelines * added git status * reverted change * added codegen options and refactoring * minor fixes' * renamed params, minor refactoring * added tests for commandline and refactoring * removed file * added back the test case * minor fixes * Update src/mlnet.Test/CommandLineTests.cs Co-Authored-By: srsaggam <41802116+srsaggam@users.noreply.github.com> * review comments * capitalize the first character * changed the name of test case * remove unused directives * Fail gracefully if unable to instantiate data view with swept parameters (#125) * gracefully fail if fail to parse a datai * rev * validate AutoFit 'Features' column must be of type R4 (#132) * Samples: exceptions / nits (#124) * Logging support in CLI + Implementation of cmd args [--name,--output,--verbosity] (#121) * addded logging and helper methods * fixing code after merge * added resx files, added logger framework, added logging messages * added new options * added spacing * minor fixes * change command description * rename option, add headers, include new param in test * formatted * build fix * changed option name * Added NlogConfig file * added back config package * fix tests * added correct validation check (#137) * Use CreateTextLoader<T>(..) instead of CreateTextLoader(..) (#138) * added support to loaddata by class in the generated code * fix tests * changed CreateTextLoader to ReadFromTextFile method. (#140) * changed textloader to readfromtextfile method * formatting * exception fixes (#136) * infer purpose of hidden columns as 'ignore' (#142) * Added approval tests and bunch of refactoring of code and normalizing namespaces (#148) * changed textloader to readfromtextfile method * formatting * added approval tests and refactoring of code * removed few comments * API 2.0 skeleton (#149) Incorporating API review feedback * The CV code should come before the training when there is no test dataset in generated code (#151) * reorder cv code * build fix * fixed structure * Format the generated code + bunch of misc tasks (#152) * added formatting and minor changes for reordering cv * fixing the template * minor changes * formatting changes * fixed approval test * removed unused nuget * added missing value replacing * added test for new transform * fix test * Update src/mlnet/Templates/Console/MLCodeGen.cs Co-Authored-By: srsaggam <41802116+srsaggam@users.noreply.github.com> * Sanitize the column names in CLI (#162) * added sanitization layer in CLI * fix test * changed exception.StackTrace to exception.ToString() * fix package name (#168) * Rev public API (#163) * Rename TransformGeneratorBase .cs to TransformGeneratorBase.cs (#153) * Fix minor version for the repository + remove Nlog config package (#171) * changed the minor version * removed the nlog config package * Added new test to columninfo and fixing up API (#178) * Make optimizing metric customizable and add trainer whitelist functionality (#172) * API rev (#181) * propagate root MLContext thru AutoML (instead of creating our own) (#182) * Enabling new command line args (#183) * fix package name * initial commit * added more commandline args * fixed tests * added headers * fix tests * fix test * rename 'AutoFitter' to 'Experiment' (#169) * added tests (#187) * rev InferColumns to accept ColumnInfo input param (#186) * Implement argument --has-header and change usage of dataset (#194) * added has header and fixed dataset and train dataset * fix tests * removed dummy command (#195) * Fix bug for regression and sanitize input label from user (#198) * removed dummy command * sanitize label and fix template * fix tests * Do not generate code concatenating columns when the dataset has a single feature column (#191) * Include some missed logging in the generated code. (#199) * added logging messages for generated code * added log messages * deleted file * cleaning up proj files (#185) * removed platform target * removed platform target * Some spaces and extra lines + bug in output path (#204) * nit picks * nit picks * fix test * accept label from user input and provide in generated code (#205) * Rev handling of weight / label columns (#203) * migrate to private ML.NET nuget for latest bug fixes (#131) * fix multiclass with nonstandard label (#207) * Multiclass nondefault label test (#208) * printing escaped chars + bug (#212) * delete unused internal samples (#211) * fix SMAC bug that causes multiclass sample to infinite loop (#209) * Rev user input validation for new API (#210) * added console message for exit and nit picks (#215) * exit when exception encountered (#216) * Seal API classes (and make EnableCaching internal) (#217) * Suggested sample nits (feel free to ask for any of these to be reverted) (#219) * User input column type validation (#218) * upgrade commandline and renaming (#221) * upgrade commandline and renaming * renaming fields * Make build.sh, init-tools.sh, & run.sh executable on OSX/Linux (#225) * CLI argument descriptions updated (#224) * CLI argument descriptions updated * No version in .csproj * added flag to disable training code (#227) * Exit if perfect model produced (#220) * removed header (#228) * removed header * added auto generated header * removed console read key (#229) * Fix model path in generated file (#230) * removed console read key * fix model path * fix test * reorder samples (#231) * remove rule that infers column purpose as categorical if # of distinct values is < 100 (#233) * Null reference exception fix for finding best model when some runs have failed (#239) * samples fixes (#238) * fix for defaulting Averaged Perceptron # of iterations to 10 (#237) * Bug bash feedback Feb 27. API changes and sample changes (#240) * Bug bash feedback Feb 27. API changes Sample changes Exception fix * Samples / API rev from 2/27 bug bash feedback (#242) * changed the directory structure for generated project (#243) * changed the directory structure for generated project * changed test * upgraded commandline package * Fix test file locations on OSX (#235) * fix test file locations on OSX * changing to Path.Combine() * Additional Path.Combine() * Remove ConsoleCodeGeneratorTests.GeneratedTrainCodeTest.received.txt * Additional Path.Combine() * add back in double comparison fix * remove metrics agent NaN returns * test fix * test format fix * mock out path Thanks to @daholste for additional fixes! * upgrade to latest ML.NET public surface (#246) * Upgrade to ML.NET 0.11 (#247) * initial changes * fix lightgbm * changed normalize method * added tests * fix tests * fix test * Private preview final API changes (#250) * .NET framework design guidelines applied to public surface * WhitelistedTrainers -> Trainers * Add estimator to public API iteration result (#248) * LightGBM pipeline serialization fix (#251) * Change order that we search for TextLoader's parameters (#256) * CLI IFileInfo null exception fix (#254) * Averaged Perceptron pipeline serialization fix (#257) * Upgrade command-line-api and default folder name change (#258) * change in defautl folderName * upgrade command line * Update src/mlnet/Program.cs Co-Authored-By: srsaggam <41802116+srsaggam@users.noreply.github.com> * eliminate IFileInfo from CLI (#260) * Rev samples towards private preview; ignored columns fix (#259) * remove unused methods in consolehelper and nit picks in generated code (#261) * nit picks * change in console helper * fix tests * add space * fix tests * added nuget sources in generated csproj (#262) * added nuget sources in csproj * changed the structure in generated code * space * upgrade to mlnet 0.11 (#263) * Formatting CLI metrics (#264) Ensures space between printed metrics (also model counter). Right aligned metrics. Extended AUC to four digits. * Add implementation of non -ova multi class trainers code gen (#267) * added non ova multi class learners * added tests * test cases * Add caching (#249) * AdvancedExperimentSettings sample nits (#265) * Add sampling key column (#268) * Initial work for multi-class classification support for CLI (#226) * Initial work for multi-class classification support for CLI * String updates * more strings * Whitelist non-OVA multi-class learners * Refactor the orchestration of AutoML calls (#272) * Do not auto-group columns with suggested purpose = 'Ignore' (#273) * Fix: during type inferencing, parse whitespace strings as NaN (#271) * Printing additional metrics in CLI for binary classification (#274) * Printing additional metrics in CLI for binary classification * Update src/mlnet/Utilities/ConsolePrinter.cs * Add API option to store models on disk (instead of in memory); fix IEstimator memory leak (#269) * Print failed iterations in CLI (#275) * change the type to float from double (#277) * cache arg implementation in CLI (#280) * cache implementation * corrected the null case * added tests for all cases * Remove duplicate value-to-key mapping transform for multiclass string labels (#283) * Add post-trainer transform SDK infra; add KeyToValueMapping transform to CLI; fix: for generated multiclass models, convert predicted label from key to original label column type (#286) * Implement ignore columns command line arg (#290) * normalize line endings * added --ignore-columns * null checks * unit tests * Print winning iteration and runtime in CLI (#288) * Print best metric and runtime * Print best metric and runtime * Line endings in AutoMLEngine.cs * Rename time column to duration to match Python SDK * Revert to MicroAccuracy and MacroAccuracy spellings * Revert spelling of BinaryClassificationMetricsAgent to BinaryMetricsAgent to reduce merge conflicts * Revert spelling of MulticlassMetricsAgent to MultiMetricsAgent to reduce merge conflicts * missed some files * Fix merge conflict * Update AutoMLEngine.cs * Add MacOS & Linux to CI; MacOS & Linux test fixes (#293) * MicroAccuracy as default for multi-class (#295) Change default optimization metric for multi-class classification to MicroAccuracy (accuracy). Previously it was set to MacroAccuracy. * Null exception for ignorecolumns in CLI (#294) * Null exception for ignorecolumns in CLI * Check if ignore-columns array has values (as the default is now a empty array) * Emit caching flag in pipeline object model. (Includes SuggestedPipelineBuilder refactor & debug string fixes / refactor) (#296) * removed sln (#297) * Caching enabling in code gen part -2 (#298) * add * added caching codegen * support comma separated values for --ignore-columns (#300) * default initialization for ignore columns (#302) * default initialization * adde null check * Codegen for multiclass non-ova (#303) * changes to template * multicalss codegen * test cases * fix test cases * Generated Project new structure. (#305) * added new templates * writing files to disck * change path * added new templates * misisng braces * fix bugs * format code * added util methods for solution file creation and addition of projects to it * added extra packages to project files * new tests * added correct path for sln * build fix * fix build * include using system in prediction class (#307) * added using * fix test * Random number generator is not thread safe (#310) * Random number generator is not thread safe * Another local random generator * Missed a few references * Referncing AutoMlUtils.random instead of a local RNG * More refs to mail RNG; remove Float as per #1669 * Missed Random.cs * Fix multiclass code gen (#314) * compile error in codegen * removes scores printing * fix bugs * fix test * Fix compile error in codegen project (#319) * removed redundant code * fix test case * Rev OVA pipeline node SDK output: wrap binary trainers as children inside parent OVA node (#317) * Ova Multi class codegen support (#321) * dummy * multiova implementation * fix tests * remove inclusion list * fix tests and console helper * Rev run result trainer name for OVA: output different trainer name for each OVA + binary learner combination (#322) * Rev run result trainer name for Ova: output different trainer name for each Ova + binary learner combination * test fixes * Console helper bug in generated code for multiclass (#323) * fix * fix test * looping perlogclass * fix test * Initial version of Progress bar impl and CLI UI experience (#325) * progressbar * added progressbar and refactoring * reverted * revert sign assembly * added headers and removed exception rethrow * Setting model directory to temp directory (#327) * Suggested changes to progress bar (#335) * progressbar * added progressbar and refactoring * reverted * revert sign assembly * added headers and removed exception rethrow * bug fixes and updates to UI * added friendly name printing for metric * formatting * Rev Samples (#334) * Telemetry2 (#333) * Create test.txt * Create test.txt * changes needed for benchmarking * forgot one file * merge conflict fix * fix build break * back out my version of the fix for Label column issue and fix the original fix * bogus file removal * undo SuggestedPipeline change * remove labelCol from pipeline suggester * fix build break * rename AutoML to Microsoft.ML.Auto everywhere and a shot at publishing nuget package (will probably need tweaks once I try to use the pipleline) * tweak queue in vsts-ci.yml * CLI telemetry implementation * Telemetry implementation * delete unnecessary file and change file size bucket to actually log log2 instead of nearest ceil value * add headers, remove comments * one more header missing * Fix progress bar in linux/osx (#336) * progressbar * added progressbar and refactoring * reverted * revert sign assembly * added headers and removed exception rethrow * bug fixes and updates to UI * added friendly name printing for metric * formatting * change from task to thread * Update src/mlnet/CodeGenerator/CodeGenerationHelper.cs Co-Authored-By: srsaggam <41802116+srsaggam@users.noreply.github.com> * Mem leak fix (#328) * Create test.txt * Create test.txt * changes needed for benchmarking * forgot one file * merge conflict fix * fix build break * back out my version of the fix for Label column issue and fix the original fix * bogus file removal * undo SuggestedPipeline change * remove labelCol from pipeline suggester * fix build break * rename AutoML to Microsoft.ML.Auto everywhere and a shot at publishing nuget package (will probably need tweaks once I try to use the pipleline) * tweak queue in vsts-ci.yml * there is still investigation to be done but this fix works and solves memory leak problems * minor refactor * Upgrade ML.NET package (#343) * Add cross-validation (CV), and auto-CV for small datasets; push common API experiment methods into base class (#287) * restore old yml for internal pipeline so we can publish nuget again to devdiv stream (#344) * Polishing the CLI UI part-1 (#338) * formatting of pbar message * Polishing the UI * optimization * rename variable * Update src/mlnet/AutoML/AutoMLEngine.cs Co-Authored-By: srsaggam <41802116+srsaggam@users.noreply.github.com> * Update src/mlnet/CodeGenerator/CodeGenerationHelper.cs Co-Authored-By: srsaggam <41802116+srsaggam@users.noreply.github.com> * new message * changed hhtp to https * added iteration num + 1 * change string name and add color to artifacts * change the message * build errors * added null checks * added exception messsages to log file * added exception messsages to log file * CLI ML.NET version upgrade (#345) * Sample revs; ColumnInformation property name revs; pre-featurizer fixes (#346) * CLI -- consume logs from AutoML SDK (#349) * Rename RunDetails --> RunDetail (#350) * command line api upgrade and progress bar rendering bug (#366) * added fix for all platforms progress bar * upgrade nuget * removed args from writeline * change in the version (#368) * fix few bugs in progressbar and verbosity (#374) * fix few bugs in progressbar and verbosity * removed unused name space * Fix for folders with space in it while generating project (#376) * support for folders with spaces * added support for paths with space * revert file * change name of var * remove spaces * SMAC fix for minimizing metrics (#363) * Formatting Regression metrics and progress bar display days. (#379) * added progress bar day display and fix regression metrics * fix formatting * added total time * formatted total time * change command name and add pbar message (#380) * change command name and add pbar message * fix tests * added aliases * duplicate alias * added another alias for task * UI missing features (#382) * added formatting changes * added accuracy specifically * downgrade the codepages (#384) * Change in project structure (#385) * initial changes * Change in project structure * correcting test * change variable name * fix tests * fix tests * fix more tests * fix codegen errors * adde log file message * changed name of args * change variable names * fix test * FileSizeBuckets in correct units (#387) * Minor telemetry change to log in correct units and make our life easier in the future * Use Ceiling instead of Round * changed order (#388) * prep work to transfer to ml.net (#389) * move test projects to top level test subdir * rename some projects to make naming consistent and make it build again * fix test project refs * Add AutoML components to build, fix issues related to that so it builds

Support for LSTM / RNN (Time Series) #149

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions