In [9]:

    

%matplotlib inline
from matplotlib import pylab as pl
import cPickle as pickle
import pandas as pd
import numpy as np
import os


    

In [10]:

    

df = pd.DataFrame()
# 0.75037
df['rf'] = pd.read_csv('../submissions/140926-predict.2.csv', index_col='clip', squeeze=True)
w_rf = 0.8

# 0.73188
df['gb'] = pd.read_csv('../submissions/140928-predict.1.csv', index_col='clip', squeeze=True)
w_gb  = 0.65

# 0.71603
df['dbn'] = pd.read_csv('../submissions/140930-predict.5.csv', index_col='clip', squeeze=True)
w_dbn = 0.45

# 0.71358
df['rfpca'] = pd.read_csv('../submissions/141001-predict.1.csv', index_col='clip', squeeze=True)
w_rfpca = 0.4

# 0.70404
df['rfica'] = pd.read_csv('../submissions/141022-predict.3.csv', index_col='clip', squeeze=True)
w_rfica = 0.35

# 0.80999
df['best'] = pd.read_csv('../submissions/141005-predict.2.csv', index_col='clip', squeeze=True)


    

In [11]:

    

pd.scatter_matrix(df[['gb','rf','rfpca','rfica','dbn','best']],figsize=(6, 6), diagonal='kde');


    

In [12]:

    

s = w_gb + w_rf + w_rfpca + w_rfica + w_dbn

w_gb /= s
w_rf /= s
w_rfpca /= s
w_rfica /= s
w_dbn /= s


    

In [13]:

    

w_gb, w_rf, w_rfpca, w_rfica, w_dbn


    

    
Out[13]:





(0.2452830188679245,
 0.3018867924528302,
 0.1509433962264151,
 0.13207547169811318,
 0.16981132075471697)

In [14]:

    

df['preictal'] = w_gb * df['gb'] + w_rf * df['rf'] + w_rfpca * df['rfpca'] + w_rfica * df['rfica'] + w_dbn * df['dbn']


    

In [15]:

    

df['preictal'].to_csv('../submissions/141022-predict.5.csv', header=True)


    

In [16]:

    

pd.scatter_matrix(df[['best','preictal']]);


    

In [8]: