charanhu

#so are we are imputing zero for missing values in target feature:

train_df['totals.transactionRevenue'].fillna(0,inplace=True)

charanhu

train_df['device.isMobile'] = train_df['device.isMobile'].astype(bool)

test_df['device.isMobile']  = test_df['device.isMobile'].astype(bool)

charanhu

numeric_feat = ['visitNumber',\
                'visitStartTime',\
                'totals.hits',\
                'totals.pageviews',\
                'totals.timeOnSite',\
                'totals.transactions',\
                'totals.transactionRevenue']   
for col in numeric_feat:
    
    train_df[col].fillna(0,inplace=True)

charanhu

# source : https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.LabelEncoder.html

categorical_feat = ['channelGrouping',\
                    'device.browser',\
                    'device.operatingSystem',\
                    'device.deviceCategory',\
                    'geoNetwork.continent',\
                    'geoNetwork.subContinent',\
                    'geoNetwork.country',\
                    'geoNetwork.region',\

charanhu

def get_time_series_features(data,k):
    
    train_frame_k = data.loc[(data['date'] >= min(data['date']) + timedelta(days=168*(k-1))) & (data['date'] < min(data['date'])  + timedelta(days=168*k)) ]
    
    # for train_frame_1 = aug 1st 2016 to jan 15 th 2017
    #in code k=1 and min(data['date']) = aug 1st 2016
    # so it return from aug 1st 2016 to jan 15 th 2017 (since: aug 1st 2016+168)
    
    test_frame_k = data.loc[(data['date'] >= max(train_frame_k['date']) + timedelta(days=46)) & (data['date'] <= max(train_frame_k['date']) + timedelta(days=108))]
    

charanhu

train_test_data = pd.concat([train_df, test_df], axis=0).reset_index()

%time train_frame_1 = get_time_series_features(train_test_data,1)
train_frame_1.to_pickle('train_frame_1')

%time train_frame_2 = get_time_series_features(train_test_data,2)
train_frame_2.to_pickle('train_frame_2')

%time train_frame_3 = get_time_series_features(train_test_data,3)
train_frame_3.to_pickle('train_frame_3')

charanhu

gridParams = {
    'learning_rate': [0.005,0.01,0.015],    #Learning rate
    'n_estimators': [40,100,200],           #number of boosting iterations  
    'num_leaves': [6,8,12,15,16],           #number of leaves in full tree
    'boosting_type' : ['gbdt'],
    'objective' : ['binary'],               #Binary Classification model to predict whether customer will return during test window
    'metric' : ['binary_logloss'],          #Performance metric as "Binary Logloss"  
    'colsample_bytree' : [0.6, 0.8, 1],     #LightGBM will select 80% of features before training each tree
    'subsample' : [0.7,0.9, 1],             #this will randomly select part of data without resampling
    'reg_alpha' : [0,1],                    #L1 regularization

charanhu

#Running Lightgbm model for 10 iterations and took average of those.
#Source :- https://www.kaggle.com/kostoglot/winning-solution

pr_lgb_sum = 0    #Variable to store predictions.

print('Training and predictions')

for i in range(10):     #Running the model for 10 iterations and would be taking average of those as final value.
    
    print('Interation number ', i)

charanhu

gridParams = {
    
    "n_estimators":[200,400,600,800,1000],
              
    "max_depth": [2,5,7,9,10],
              
    "min_samples_split":[2,3,5,7],
              
    "min_samples_leaf": [1,2,4]
             

charanhu

gridParams = {
    'learning_rate': [0.005,0.01,0.015],    #Learning rate
    'n_estimators': [40,100,200],           #number of boosting iterations  
    'num_leaves': [6,8,12,15,16],           #number of leaves in full tree
    'boosting_type' : ['gbdt'],
    'objective' : ['binary'],               #Binary Classification model to predict whether customer will return during test window
    'metric' : ['binary_logloss'],          #Performance metric as "Binary Logloss"  
    'colsample_bytree' : [0.6, 0.8, 1],     #LightGBM will select 80% of features before training each tree
    'subsample' : [0.7,0.9, 1],             #this will randomly select part of data without resampling
    'reg_alpha' : [0,1],                    #L1 regularization