* adding supun's polyfeatures code
* interaction_only flag breaks this, commenting out for now
* removing extra dividers
Co-authored-by: Matteo Interlandi <m.interlandi@gmail.com>
* adding Supun's missing indicator and imputer code
* adding transform flag to classes
* adding tests. TODO: problem with missing indicator
* rolling back missing indicator for now
* cleaning up test
* combinging simpleimputer and missingindicator into one imputer file
* deleting old html
* fixing stale comment in test file
* undoing overzealous pdoc
* refactoring of containers before adding torchscript support
* add torchscript backend
* add tests
* add device as a parameter to convert, add 'to' to the pytorch containers
* improve coverage
* update doc
* add automatic doc generation to pre-commit
* add templates and pdoc to pre-commit
* make pdoc non-blocking
* remove nonblocking
Co-authored-by: Karla Saur <1703543+ksaur@users.noreply.github.com>
* rename calibration in post transform
Make post transform configurable
* done with regression
fixed problem with post_transform not being called
* working on testing
* fix for decision tree binary classification
* compute probabilities for sklearn models directly during conversion
* Fix binary and multi for decision trees
* Fix problem with probabilities for sklearn models
Fix problem with onnx tree models with only 1 node.
* post_transform is only executed for gbdt models
* merged with master
* rename sv test file
* update to the documentation
* Addressing Karla's comments.
* bump pytorch version
* get the correct version of torch for mac and linux
* fix pytorch version for python 3.5 to 1.5.1
* disable gemm for onnx
* just use tree_trav as default
* getting started with OHE onnx
* blocked on ArrayFeatureExtractor
* getting started with OHE onnx
* Fix bug in supported
add initializers to inputs
* add concat
* add reshape
* fix the constant that got removed
* add missing start to cast
* fix onnx types
* commenting out strings for now, issue filed
* simplifying failure case test
* removing redundant line
* adding skipif
Co-authored-by: Matteo Interlandi <mainterl@microsoft.com>
* VarianceThreshold passing, need to rework the tests/code for SelectPercentile, SelectKBest
* copypaste err
* removing SelectPerc for now until a solution can be found. Adding placeholder onnxml AFE code to be used with OHE PR
* matteo's comments
* matteo's comments
* Add float64 data tests
* Add support for float64 data
* float64 data support: Add regressor tests
* Update blog example
* Fix blogpost nb
* Update sklearn random forest nb
* XGBoost nb update
* Update lightgbm nb
* Update nbs
* Add lightgbm tests
* Fix sklearn test s
* Add additional tests around float64 data
* Remove onnx runtime based tests
* Add additional float64 data tests
* Update README example
* Skip xgboost tests when it is not installed
* Fix formatting issues
* nit
* pulling up Scalar into impl file
* first stab at onnx scaler conv
* standardscaler test
* preparing to add other types of scalers
* adding the rest of the scaler tests
* matteo's comments
* placeholders
* todo, clarify why onnxml logistic regression name doesn't line up w SKL
* tests for regressor
* adding test for parsefail
* addressing matteo's feedback
* adding back comment
* add converter for isolation forest and add is_anomaly_detection to base tree implementation classes
* add PyTorchBackendModelAnomalyDetection to container and topology
* add constants and support list
* add tests for isolation forest converter
* fix flake 8 checks
* make PyTorchBackendModelAnomalyDetection inherit from regression
* refactor docstring for _average_path_length
* add kwargs in base implementation classes
* fix style errors
Co-authored-by: Zhanjie Zhu <zhanjie.zhu@ing.com>
Co-authored-by: Matteo Interlandi <mainterl@microsoft.com>
* fixing typo in tree
* renaming onnxml tree test file to match others
* pulling LinearModel class up into _linear_impl
* fixing typo in tree
* renaming onnxml tree test file to match others
* fixing onnx_installed ->onnx_runtime_installed
* docstring, removing skl2onnx
* stuck at Inference of ONNX models with multiple ONNXML operators is not supported yet.
* commenting tests out for now
Matteo Interlandi
Karla Saur
Supun Nakandala
kernc
KranthiGV
zz