Self-writing library for more best practice when participating in competitions at Kaggle.com.
Wrote for myself.

print_bold(string)

Function to display in output JupyterNotebook markdowns

submit_result(df,id,target,path,name,score = 0,oof = None)

Function to from submit at kaggle.
Create two folder in path folder, oof - with predict at train and submition - with predict at test.

• df - pandas DataFrame with id and target
• id - id field in df
• target - target feature in df
• path - path to save submit on local machine
• name - name for file with submit
• score - score at CV if exist
• oof - pandas DataFrame with predict at train if exist

smoothed_aggregate(df, null_field, agg_field, alpha = 10)

Smoothed aggregate, that use to reduce overfiting

• df - pd.DataFrame() with null_field and agg_field

• null_field - field, that will be used in groupby

• agg_fueld - field, that need be aggregate

• alpha - coefficent for smooth

• return - result pd.Series() with aggregated values

def reduce_mem_usage(df, verbose=True, less_data = True)

Compresse DataFrame for low mem usage.
!!!WARNING!!! The default parameter less_data = True, that mean,
while you use this function, tou understand that while you compress
float value you may lose precision in decimal places.
If you don't want it - set parameter less_data to False

def ensemble_predictions(predictions, weights=None, type_="linear")

Function to ansamble prediction.

• predictions - array with predictions

• weights - weight for prediction in array

• type_ - tpe of mix

  • 'linear' - simple mean stuck
  • 'harmonic' - ?
  • 'geometric' - ?
  • 'rank' - ranked stuck (vote method)

• return result of stuck (array)

def lgbm_calc(train,
              test,
              features,
              target,
              param,
              score_function = roc_auc_score,
              n_fold = 3, 
              seed = 11, 
              cat_features = []
              ):

Function for predicting with lgbm, that use KFold method

• train - train dataset
• test - test dataset
• features - features, that will be used for predict
• target - target feature
• param - param for LGBM (see doc. for lgbm)
• score_function - score function from sklearn.metrics or you own function
• n_fold - number folds for KFold
• seed - seed for random
• cat_features - categorical feature if tou have it in dataset (default [])

return :

• oof_df - dataframe with predict for train part
• submit - dataframe with predict for test part
• fi - feature importance of training

def catboost_calc(train, test, features, target, param, score_function = roc_auc_score, n_fold = 3, seed = 11, cat_features = [] ): Function for predicting with CatBoost(Yandex), that use KFold method

train - train dataset test - test dataset features - features, that will be used for predict target - target feature param - param for LGBM (see doc. for lgbm) score_function - score function from sklearn.metrics or you own function n_fold - number folds for KFold seed - seed for random cat_features - categorical feature if tou have it in dataset (default []) return :

oof_df - dataframe with predict for train part submit - dataframe with predict for test part