In [1]:

    

import numpy as np
import pandas as pd
from sklearn.ensemble import GradientBoostingClassifier, RandomForestClassifier, AdaBoostClassifier, BaggingClassifier
import datetime


    

In [2]:

    

def now():
    return '_'.join(str(datetime.datetime.now()).split())


    

In [3]:

    

df_train = pd.read_csv('raw_data/train.csv')
df_test = pd.read_csv('raw_data/test.csv')


    

In [4]:

    

def convert_class_label_to_int(class_label):
    return int(class_label[6:]) - 1


    

In [5]:

    

def find_means_stds(features_array):
    means = [np.mean(features_array[:, i]) for i in range(features_array.shape[1])]
    stds = [np.std(features_array[:, i]) for i in range(features_array.shape[1])]
    return {'means': means, 'stds': stds}


    

In [6]:

    

def z_score_feature(feature_slice, mean, std):
    return (feature_slice - mean) / std


    

In [7]:

    

renamed_labels = [convert_class_label_to_int(i) for i in df_train['target'].values]

feature_columns = ['feat_' + str(i + 1) for i in range(93)]
df_train['renamed_labels'] = renamed_labels


    

In [8]:

    

mstddict = find_means_stds(df_train[feature_columns].values)


    

In [9]:

    

for i in range(93):
    df_train['feat_' + str(i + 1)] = z_score_feature(df_train['feat_' + str(i + 1)].values, mstddict['means'][i],
                                                     mstddict['stds'][i])


    

In [10]:

    

for i in range(93):
    df_test['feat_' + str(i + 1)] = z_score_feature(df_test['feat_' + str(i + 1)].values, mstddict['means'][i],
                                                    mstddict['stds'][i])


    

In [11]:

    

# rf = AdaBoostClassifier(n_estimators=100)
# clrf = RandomForestClassifier(n_estimators=400, max_features=25)
# clf = AdaBoostClassifier(base_estimator=clrf, n_estimators=15)
# clf2 = GradientBoostingClassifier(n_estimators=250, max_depth=8, max_features=15)

clf = GradientBoostingClassifier(n_estimators=700, max_depth=7, max_features=20, learning_rate=0.03)


    

In [12]:

    

clf.fit(df_train[feature_columns].values, df_train['renamed_labels'].values)


    

    
Out[12]:





GradientBoostingClassifier(init=None, learning_rate=0.03, loss='deviance',
              max_depth=7, max_features=20, max_leaf_nodes=None,
              min_samples_leaf=1, min_samples_split=2,
              min_weight_fraction_leaf=0.0, n_estimators=700,
              random_state=None, subsample=1.0, verbose=0,
              warm_start=False)

In [13]:

    

labels_pr = clf.predict_proba(df_test[feature_columns].values)
labels_pr_tr = clf.predict_proba(df_train[feature_columns].values)


    

In [14]:

    

now_name = now()

predict_dict = {'id': df_test['id'].values}
for i in range(9):
    predict_dict['Class_' + str(i + 1)] = labels_pr[:, i]
df_sub = pd.DataFrame(predict_dict)

fname = 'submissions/' + now_name + '.csv'
df_sub.to_csv(fname, index=False)

predict_dict = {'id': df_train['id'].values}
for i in range(9):
    predict_dict['Class_' + str(i + 1)] = labels_pr_tr[:, i]
df_sub = pd.DataFrame(predict_dict)
fname = 'train_predict/' + now_name + '.csv'
df_sub.to_csv(fname, index=False)


    

In [ ]: