serg-delft / dslinter Goto Github PK

refactor UnnecessaryIterationTensorflowChecker and TestMemoryReleaseTensorflowChecker.

Implement the DependentThresholdChecker for pytorch.

Implement the DataframeCoversionChecker for pandas

Check random seed setting in tensorflow.

Implement TensorArrayChecker for tensorflow2
https://hynn01.github.io/dslinter/posts/codesmells/19-tensorarray-not-used/

Implement TrainEvalModeTogglingChecker for pytorch.
https://hynn01.github.io/dslinter/posts/codesmells/21-training-evaluation-mode-improper-toggling/

I think there is a bug in the dslinter plugin.

Minimal example: minimal-example.zip

Steps to reproduce:

• Unzip the minimal example
• cd into the dslinter-test directory
• run pipenv install --dev to install dslinter
• run pipenv run pylint src --load-plugins dslinter --output-format=json:result.json

Expected beviour:

• Two code smells written to the result.json file

Actual behaviour:

• No content in the result.json file.

If you comment out the from os import path line it does write to the result.json file, so that is likely the cause. It's also not only the json format, but also the text format that shows this behaviour. When not enabling the dslinter plugin, it does write to the files.

Running it with the command of the steps to follow document does work, but we want to use all the checkers of pylint.

• https://hynn01.github.io/ml-smells/posts/codesmells/21-data-leakage/
• The checker's rule: If the variable in the train_test_split function is ever assigned to a preprocessing function with fit() or fit_transform() call, it's possible that there is a data leakage.

Check whether random seed is set in pandas.

Refactor the ScalerChecker (check whether these is no scaling before scaling-sesitive operation) for scikit-learn.

Implement LossAPIChecker for pytorch
https://hynn01.github.io/dslinter/posts/codesmells/23-loss-api-misused/

Implement DataLoaderUsingChecker for pytorch
https://hynn01.github.io/dslinter/posts/codesmells/24-dataloader-unused/

Implement the ChainIndexingChecker for pandas.

Implement the ColumnChecker for Pandas.
https://hynn01.github.io/dslinter/posts/codesmells/15-columns-and-datatype-not-explicitly-set/

Implement Broadcasting feature for tensorflow2.
https://hynn01.github.io/dslinter/posts/codesmells/18-broadcasting-feature-not-used/

• Since data-science-types stopped development, using mypy and data-science-types to infer pandas's type can sometimes cause error.
• Currently inplace-pandas checker and unneccessary-iteration-pandas still use mypy and data-science-types to infer pandas's type.
• We should probably look for a better solution for infering data science libraries' types.

Explore whether Data Leakage Checker can be implemented for pytorch and tensorflow.

Check random seed setting in scikit-learn.

Explore whether ScalerChecker (check whether these is no scaling before scaling-sesitive operation) can be applied to pytorch and tensorflow

Explore which hyperparameters should be explicitly checked in pytorch and tensorflow.

Implement GradientClearChecker for pytorch.
https://hynn01.github.io/dslinter/posts/codesmells/22-gradients-not-cleared-before-backward-propagation/

Implement MergeParameterChecker for pandas.
https://hynn01.github.io/dslinter/posts/codesmells/17-merge-api-parameter-not-explicitly-set/

Implement DeterministicAlgorithmChecker for pytorch

Check whether random seed is set in numpy.

Hi!

I have recently tried to install dslinter in a project where I was also using a version of pytest > 3.10.1 and pip install -r requirements.txt errors out because of dependency conflicts between the two.

As far as I can tell, pytest is used in dslinter purely as a testing library, and is not required for actually running the tool. Thus, would it be possible to remove pytest (and, possibly, pytest-cov) from pyproject.toml, such that other projects using dslinter are not limited in their choice of pytest version?

Implement the MatrixMultiplicatioChecker for numpy.
https://hynn01.github.io/dslinter/posts/codesmells/14-matrix-multiplication-api-misused/

Implement NetForwardChecker for pytorch.
https://hynn01.github.io/dslinter/posts/codesmells/20-pytorch-call-method-misused/

Implement PossibleInvalidValueChecker(check whether there is a possible invalid value but there is no mask for it) for tensorflow.

Implement the DataType Checker for pandas.
https://hynn01.github.io/dslinter/posts/codesmells/15-columns-and-datatype-not-explicitly-set/

Explore whether PossibleInvalidValueChecker can be implemented for pytorch

Implement the DependentThresholdChecker for tensorflow.

Implement the ColumnInitialization checker for pandas.
https://hynn01.github.io/dslinter/posts/codesmells/16-empty-column-misinitialization/

Check whether random seed is set in pytorch.