kowalsky 0.0.38

A small package for all useful ML things

Kowalsky, analysis!

A simple package for handful ML things and more.

What's new? [v0.0.38]

• add feature package with two types of analysis + support для остальных функций
  • Recursive Feature Elimination
  • Sequential Feature Selection
• improve optimize:
  • EarlyStopping mechanism
  • optimization graph
  • multitasks with n_jobs=-1
• add logs package

What's inside?

1. analysis - method for evaluation of specified model with given dataframe. With export_test_set=True it exports ready for submission predictions.

2. df - module for working with dataframe:

   • corr - sort all correlated features.
   • handle_outliers - fill or drop columns with outliers.
   • log_transform - transform columns with log function.
   • group_by_mean - make additional columns with aggregated mean
   • group_by_max - make additional columns with aggregated max
   • group_by_min - make additional columns with aggregated min
   • apply_with_progress - apply heavy function for each row of dataset.
   • scale - scale columns with Standard of MinMax scalers

3. kaggle:

   • submit - make submit-file for kaggle based on sample

4. logs:

   • profile_memory - logs all heavy variables
   • make_pretty_pyplot - makes pyplot look better :)

5. optuna - handful methods for working with optuna:

   • optimize - optimize model with given dataframe
   • optimize_super_learner - optimize super learner configuration with given set of models and set of heads (meta_model)

6. colab:

   • csv - read csv file located at Google Drive with specified id
   • path - get path to Google Drive file

7. feature:

   • rfe_analysis - Recursive Feature Elimination analysis
   • sfs_analysis - Sequential Feature Selection analysis

What's next?

• Use optuna for searching the best feature amount
• Add file logger to track the progress in JupterLab

Example:

!pip install kowalsky --upgrade
from kowalsky.optuna import optimize
optimize('RFR',
         path='../input/project/feed.csv',
         scorer='acc',
         y_label='y_label',
         trials=3000)

Avaliable models:

Gradient Boosts

    'xgbR': XGBRegressor
    'xgbC': XGBClassifier
    'lgbR': LGBMRegressor
    'lgbC': LGBMClassifier

Trees

    'rfR': RandomForestRegressor
    'rfC': RandomForestClassifier
    'dtR': DecisionTreeRegressor
    'dtC': DecisionTreeClassifier
    'etR': ExtraTreeRegressor
    'etC': ExtraTreeClassifier

Ensemble

    'baggC': BaggingClassifier
    'baggR': BaggingRegressor
    'adaR': AdaBoostRegressor
    'adaC': AdaBoostClassifier
    'cbR': CatBoostRegressor
    'cbC': CatBoostClassifier

KNeighbors

    'knC': KNeighborsClassifier
    'knR': KNeighborsRegressor

SVM

    'svR': SVR
    'svC': SVC