# Bimbo Python XGBoost 5 week Lag Python script adapted from R # Rodolfo Lomascolo e-mail: r.lomascolo@rodolfolomascolo.com # Adapted from code by Bohdan Pavlyshenko http://tinyurl.com/jd6k2kr # All credits for Bohdan Pavlyshenko and all bugs to me. # # Due to heavy usage of memory, this script will not run in kaggle. # the script runs only with a minimum of 35GB of physical or physical+swap memory # it can take a couple of hours to train depending on your machine, disk and memory # I used an 8 core, 56GB of RAM Azure DS13 machine with additional 100 GB of swap import numpy as np # linear algebra import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv) from subprocess import check_output from sklearn.metrics import make_scorer, mean_squared_error from sklearn.cross_validation import train_test_split from sklearn import preprocessing import xgboost as xgb from sklearn import cross_validation import scipy.stats as stats import gc train = pd.read_csv('../input/train.csv') test = pd.read_csv('../input/test.csv') print('Train and Test Read') train['target'] = train['Demanda_uni_equil'] train.drop(['Demanda_uni_equil'],axis=1, inplace = True) train['tst'] = 0 test['tst'] = 1 data = pd.concat([train,test], axis=0, copy=True) print('Train and Test Concat') del train del test gc.collect() for i in range(1,6): lag = 'Lag' + str(i) print('Lag:',lag) data1 = data[['Semana','Cliente_ID','Producto_ID','target']] data1.loc[:,'Semana'] = data1['Semana'] +i data1 = pd.groupby(data1,['Semana','Cliente_ID','Producto_ID']).mean() data1 = data1.reset_index() data1.rename(columns={'target': lag}, inplace=True) data = pd.merge(data,data1, how='left', left_on=['Semana','Cliente_ID','Producto_ID'], right_on=['Semana','Cliente_ID','Producto_ID'], left_index=False, right_index=False, sort=True, suffixes=('_x', '_y'), copy=False, ) del data1 gc.collect() data['TotalLags'] = data['Lag1'] + data['Lag2']+ data['Lag3']+ data['Lag4']+ data['Lag5'] data=data[data['Semana']>8] # NOW I WORK WITH WEEKS 9,10,11 nAgencia = pd.DataFrame(pd.groupby(data,['Agencia_ID','Semana'])['target'].count()) nAgencia = nAgencia.reset_index() nAgencia.rename(columns={'target': 'nAgencia'}, inplace=True) nAgencia = pd.DataFrame(pd.groupby(nAgencia,['Agencia_ID'])['nAgencia'].mean()) nAgencia = nAgencia.reset_index() data = pd.merge(data, nAgencia, how='left', left_on=['Agencia_ID'], right_on=['Agencia_ID'], left_index=False, right_index=False, sort=True, suffixes=('_x', '_y'), copy=False) del nAgencia gc.collect() print('merge completo nAgencia') print(data.shape[0]) nRuta_SAK = pd.DataFrame(pd.groupby(data,['Ruta_SAK','Semana'])['target'].count()) nRuta_SAK = nRuta_SAK.reset_index() nRuta_SAK.rename(columns={'target': 'nRuta_SAK'}, inplace=True) nRuta_SAK = pd.DataFrame(pd.groupby(nRuta_SAK,['Ruta_SAK'])['nRuta_SAK'].mean()) nRuta_SAK = nRuta_SAK.reset_index() data = pd.merge(data, nRuta_SAK, how='left', left_on=['Ruta_SAK'], right_on=['Ruta_SAK'], left_index=False, right_index=False, sort=True, suffixes=('_x', '_y'), copy=False) del nRuta_SAK gc.collect() print('merge completo nRuta_SAK') print(data.shape[0]) nCliente_ID = pd.DataFrame(pd.groupby(data,['Cliente_ID','Semana'])['target'].count()) nCliente_ID = nCliente_ID.reset_index() nCliente_ID.rename(columns={'target': 'nCliente_ID'}, inplace=True) nCliente_ID = pd.DataFrame(pd.groupby(nCliente_ID,['Cliente_ID'])['nCliente_ID'].mean()) nCliente_ID = nCliente_ID.reset_index() data = pd.merge(data, nCliente_ID, how='left', left_on=['Cliente_ID'], right_on=['Cliente_ID'], left_index=False, right_index=False, sort=True, suffixes=('_x', '_y'), copy=False) del nCliente_ID gc.collect() print('merge completo nCliente_ID') print(data.shape[0]) nProducto_ID = pd.DataFrame(pd.groupby(data,['Producto_ID','Semana'])['target'].count()) nProducto_ID = nProducto_ID.reset_index() nProducto_ID.rename(columns={'target': 'nProducto_ID'}, inplace=True) nProducto_ID = pd.DataFrame(pd.groupby(nProducto_ID,['Producto_ID'])['nProducto_ID'].mean()) nProducto_ID = nProducto_ID.reset_index() data = pd.merge(data, nProducto_ID, how='left', left_on=['Producto_ID'], right_on=['Producto_ID'], left_index=False, right_index=False, sort=True, suffixes=('_x', '_y'), copy=False) del nProducto_ID gc.collect() print('merge completo nProducto_ID') print(data.shape[0]) data.replace(np.nan,0, inplace=True) train = data[data['tst']==0] predict = data[data['tst']==1] train['target'] = np.log(train['target'] + 1) #train2 = train.sample(n=1000000) <-- another possible reduction of data for fast testing train2 = train y = train['target'] X = train[[ 'Agencia_ID','Canal_ID','Cliente_ID','Producto_ID','Ruta_SAK', 'Lag1','Lag2','Lag3','Lag4','Lag5','TotalLags', 'nAgencia','nRuta_SAK','nCliente_ID','nProducto_ID']] print(X.shape, y.shape) # split into train and test X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.01, random_state=1729) print(X_train.shape, X_test.shape) xlf = xgb.XGBRegressor(max_depth=10, learning_rate=0.1, n_estimators=10, silent=True, objective='reg:linear', nthread=-1, gamma=0, min_child_weight=1, max_delta_step=0, subsample=0.85, colsample_bytree=0.7, colsample_bylevel=1, reg_alpha=0, reg_lambda=1, scale_pos_weight=1, seed=1440, missing=None) xlf.fit(X_train, y_train, eval_metric='rmse', verbose = True, eval_set = [(X_test, y_test)],early_stopping_rounds=100) # calculate the auc score preds = xlf.predict(X_test) print('\nRoot Mean Square error of Log(Demanda) X_test" ', mean_squared_error(y_test,preds)**0.5) # GENERATE PREDICTION FOR 10-S11 FOR KAGGLE SUBMISSION print('Make prediction for the 10th week') data_test1=predict[predict['Semana']==10] ids_10 =data_test1['id'] data_test1 = data_test1[['Agencia_ID','Canal_ID','Cliente_ID','Producto_ID','Ruta_SAK', 'Lag1','Lag2','Lag3','Lag4','Lag5','TotalLags', 'nAgencia','nRuta_SAK','nCliente_ID','nProducto_ID']] pred = xlf.predict(data_test1) res=np.exp(pred)-1 # Create lagged values of target variable which will be used as a feature for the 11th week prediction print('Create lagged values of target variable which will be used as a feature for the 11th week prediction') data_test_lag1=data_test1[['Cliente_ID','Producto_ID']] data_test_lag1['targetl1']=res data_test_lag1 = pd.groupby(data_test_lag1,['Cliente_ID','Producto_ID']).mean() data_test_lag1 = data_test_lag1.reset_index() data_test_lag1.rename(columns={'targetl1': 'Lag1'}, inplace=True) # Save the predictions for first week data_test1['Demanda_uni_equil'] = res data_test1['id'] = ids_10.astype(int).tolist() # Make prediction for the 11th week print('Make prediction for the 11th week') data_test2=predict[predict['Semana']==11] ids_11 =data_test2['id'] data_test2 = data_test2[[ 'Agencia_ID','Canal_ID','Cliente_ID','Producto_ID','Ruta_SAK', 'Lag2','Lag3','Lag4','Lag5','TotalLags', 'nAgencia','nRuta_SAK','nCliente_ID','nProducto_ID']] data_test2 = pd.merge(data_test2, data_test_lag1, how='left', left_on=['Cliente_ID','Producto_ID'], right_on=['Cliente_ID','Producto_ID'], left_index=False, right_index=False, sort=True, suffixes=('_x', '_y'), copy=False) data_test2['TotalLags'] = data_test2['Lag1'] + data_test2['Lag2']+ data_test2['Lag3']+ data_test2['Lag4']+ data_test2['Lag5'] data_test2 = data_test2[['Agencia_ID','Canal_ID','Cliente_ID','Producto_ID','Ruta_SAK', 'Lag1','Lag2','Lag3','Lag4','Lag5','TotalLags', 'nAgencia','nRuta_SAK','nCliente_ID','nProducto_ID']] data_test2.replace(np.nan,0, inplace=True) pred = xlf.predict(data_test2) res=np.exp(pred)-1 # Save the predictions for Second week data_test2['Demanda_uni_equil'] = res data_test2['id'] = ids_11.astype(int).tolist() submit = pd.concat([data_test1[['id','Demanda_uni_equil']],data_test2[['id','Demanda_uni_equil']]], axis=0, copy=True) print('Saving Predictions file') submit.to_csv('Predictions/S10-S11-Prediction.csv', index=False)