Sklearn Porter Versions Save

0.7.1 and 0.7.2

These patches solve build problems in version 0.7.0.

Fixed

• Fix installation issues with the centralised meta information and the build process
• Fix missed package and add six to the requirements.txt file

0.7.0

This is a minor update before the next major release 1.0.0.

Fixed

• Fix indices format in RandomForestClassifier (#41, thanks @apasanen)

Added

• Add Python 3.7 with Xenial to CI for testing
• Add PyTest for extended testing (see pytest.ini)
• Add useful Makefile tasks with dependency handling (see Makefile):
  • install.environment to install a conda environment
  • install.requirements to install all pip requirements
  • make link to install porter (cli) to the command line
  • make open.examples to start a local jupyter server
  • make stop.examples to stop the started jupyter server
  • make test to run all unittests in tests
  • make lint to run pylint over sklearn_porter
  • make jupytext to generate the notebooks from Python sources

Changed

• Core:
  • Merge all meta information to sklearn_porter/package.json
• CI and requirements:
  • Remove conda from CI
  • Clear and split pip requirements into three parts:
    • requirements.txt for library requirements
    • requirements.examples.txt for examples requirements
    • requirements.development.txt for development requirements
  • Update extended requriements (e.g. GCC, PHP, ...)
• Utils:
  • Refactor class sklearn_porter.utils.Shell to remove redundancy
  • Refactor CLI with a bunch of improvements (grouped arguments, use of console scripts)
• Testing:
  • Change web server port for testing from 8080 to 8713 (because of collusion with the default port of Jenkins)
  • Remove class tests.utils.DependencyChecker, instead use sklearn_porter.utils.Environment
  • Remove class tests.utils.Timer, instead use --durations=0 of PyTest

Fixed

• Fix getter of transpiled estimator and remove filename manipulations (#b5efe78)
• Fix and add binary to string conversion in Python 3.x (#bdcfb1f)
• Fix and close open template file(s) (#b5efe78)

Added

• Add new estimator:
  • Go:
    • ensemble.RandomForestClassifier (PR #24 by tpotega)

Fixed

• Fix wrong method signature of the DecisionTreeClassifier in C (Issue #23 by sanjivsoni17)

Added

• Add changelog.md for the next releases (#ee44ac9).
• Add changelog.md and readme.md to each build of a release (#ee44ac9).
• Add for each target programming language a new new command-line argument (e.g. --java, --c or --go) (#41b93a0).
• Add argument --class_name and --method_name to define the class and method name in the final output directly (#6f2a1d9).
• Add pipe functionality and the related command-line argument (--pipe or -p) (#8a57746).
• Add --export argument to dump the model data and use the specific templates (#0669645).
• Add --checksum argument to append the computed md5 checksum at the end of the dumped model data file (#cd12827).
• Add --data argument to kust export the model data (#fad499a).
• Add test class Go in tests/language/Go.py to test all implementations for the target programming language Go (#1d0b5d6).
• Add Go compiling (go build -o brain brain.go) and execution (./brain) command (#5d24f57).
• Add initial Web Workers features in JavaScript templates (#87d3236) and a create seperate example (#187efac).
• Add the feature to read the estimator from a used Pipeline (#b92edff, issue: #18).
• Add a new class argument (num_format=lambda x: str(x)) to change the default representation of floating-point values (#7f9fac8).
• Use estimator name as default class name (e.g. MLPClasifier, KNeighborsClassifier, SVC, ...) (#710a854).
• Add new estimator:
  • Go:
    • tree.DecisionTreeClassifier (#fe59710)
  • JavaScript:
    • naive_bayes.BernoulliNB (#9784d6b)
  • PHP:
    • ensemble.RandomForestClassifier (#faac38d)
    • ensemble.ExtraTreesClassifier (#2a29321)
  • Ruby:
    • svm.SVC (#3ef1646)
    • svm.NuSVC (#0a39aaf)
    • tree.DecisionTreeClassifier (#a404c4f)
    • ensemble.RandomForestClassifier (#3775501)
    • ensemble.ExtraTreesClassifier (#81b9914)
• Add support to export and import model data:
  • Java:
    • svm.SVC notebook (#e753252)
    • svm.NuSVC notebook (#e753252)
    • svm.LinearSVC notebook (#562ed8c)
    • ensemble.AdaBoostClassifier notebook (#79d846f)
    • ensemble.RandomForestClassifier notebook (#9979a94)
    • ensemble.ExtraTreesClassifier notebook (#9979a94)
    • tree.DecisionTreeClassifier notebook (#9979a94)
    • neighbors.KNeighborsClassifier notebook (#1b81d3f)
    • naive_bayes.GaussianNB notebook (#b68c5df)
    • naive_bayes.BernoulliNB notebook (#7eae57d)
    • neural_network.MLPClassifier notebook (#b988f57, #d8ff774)
  • JavaScript:
    • ensemble.AdaBoostClassifier notebook (#9fdcc49, #dd70ac7, #2401fba)
    • neighbors.KNeighborsClassifier notebook (#f915b70)
    • neural_network.MLPClassifier notebook (#ce23148)
• Add new recipes:
  • JavaScript:
    • MLPClassifier Web Workers (#187efac)

Changed

• Use human-readable placeholders (e.g. '{class_name}.{method_name}') instead of index-based placeholders (e.g. '{0}.{1}') in all main templates of all estimators (#de02795).
• Change the order of optional and required arguments in the --help text (#54d9973).
• Change the default representation of floating-point values from repr(x) to str(x) (#7f9fac8).
• Use the method name integrity_score(X) instead of predict_test(X) to avoid misconceptions for the integrity test (#715ec7d).
• Separate the model data from the algorithm:
  • tree.DecisionTreeClassifier (#f669aab, #bba6296, #b727186, #e2740fd, #5c9da8a)
  • neighbors.KNeighborsClassifier (#59a0e91, #1ac5d8a)
  • neural_network.MLPClassifier (#635da46, #7d31668, #78296e2, #4cdcfde, #7820508, #7820508)
  • neural_network.MLPRegressor (#60d9d42, #e4a8169)
  • naive_bayes.GaussianNB (#1ac5d8a)
  • naive_bayes.BernoulliNB (#ff82bb8, #3c57a06)
  • svm.SVC and svm.NuSVC (#4745d8b, #5f77e4d, #59831da, #cd8f52e, #c483d25)
  • svm.LinearSVC (#bb617c7)

Removed

• Hide the command-line argument --language and -l for the choice of the target programming language (#fc14a3b).

Fixed

• Fix inaccuracies in neural_network.MLPRegressor and neural_network.MLPClassifier occurred by the transpiled tanh and identity activation functions (#6696410).
• Fix installation problems with pip and Python 3 (#2935828, issue: #17)
• Fix dynamic class name in the MLPClassifier template (#b988f57)

Bugfix:

• Add possibility using the best estimators of optimizers automatically https://github.com/nok/sklearn-porter/issues/18

Bugfix:

• Fix installation bug https://github.com/nok/sklearn-porter/issues/17

Bugfix:

• Fix voting mechanism in sklearn.svm.SVC (commit)

Algorithms:

• Add first regressor: sklearn.neural_network.MLPRegressor ✨
• Add binary classification to:
  • sklearn.svm.LinearSVC
  • sklearn.svm.SVC.html
  • sklearn.svm.NuSVC
  • sklearn.neural_network.MLPClassifier
• Add JavaScript syntax for sklearn.naive_bayes.GaussianNB

Changes:

• Refactor tests and add new generic tests
• Add breast_cancer, digits and iris dataset to all classifier tests
• Add new environment variables N_RANDOM_FEATURE_SETS and N_EXISTING_FEATURE_SETS

New features:

• Add parameter 'use_repr' in the method export to control the output of pointing-float values.

New features:

• Prediction in the target programming language from Python
• Computation of the accuracy between the ported and original estimator

Add changes:

Package

• Extend backwards compatibility (scikit-learn>=0.14.1) by refactoring the dependency handling, tests and examples

Fixed bugs:

AdaBoostClassifier, DecisionTreeClassifier & RandomForestClassifier

• Fix error with negative indices (https://github.com/nok/sklearn-porter/issues/5)