Linear Model

Mbenz Kaggle solution 0

import pandas as pd
import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.linear_model import LinearRegression
from sklearn.feature_selection import RFE
from sklearn.cross_validation import cross_val_score, train_test_split

#input the data
data0 = pd.read_csv('./data/train.csv', sep=',', header=False, names=range(378))
data = data0
#transform the categorical data into numerical integers
for col in range(2,10):
	nofc = data.shape[1]
	for i in range(len(data[col].unique())):
		data[data.shape[1]]=0
	mapdic = {}
	for i, ai in enumerate(list(data[col].unique())):
		mapdic[ai]=i
	data[col] = data[col].map(mapdic)
	for i in range(data.shape[0]):
		data[nofc+data[col][i]][i]=1
X = data
#remove the features with proportion of value 1 smaller than 0.01
predictors=range(10, 572)
for i in range(10, 572):
	if float(sum(X[i]==1))/X.shape[0]<0.01:
		predictors.remove(i)
target = 1
#feature selection RFE(), and regression modle LinearRegression()
X = np.asarray(data[predictors])
Y = np.asarray(data[target])
model = LinearRegression()
for nfeature in range(170, 171, 2):
 ssttdd = []
 # do 40 independent training in order to estimate the mean and std of the R2 score
 for ii in range(0, 40):
	rfe = RFE(model, nfeature)
	fit = rfe.fit(X, Y)
	predictor = [predictors[i] for i in range(len(fit.support_)) if fit.support_[i]]
	train, test = train_test_split(data, test_size=0.3)
	mn = train[target].mean()
	std = train[target].std()
	lr = LinearRegression()
	lr.fit(train[predictor], train[target])
	prt = lr.predict(test[predictor])
	for ii in range(len(prt)):
		if np.abs(prt[ii]-mn)>40.0*std:
			prt[ii]=mn
	ssttdd.append((prt-test[target]).std())
 print np.mean(ssttdd)
 print np.std(ssttdd)