def split_y(data, col):
replace_nan(data, col)
print data[:, col]
median = np.median(data[:, col])
print median
for i in range(data.shape[0]):
print '========='
print data[i,[3,7,8]]
print data[i, col]
if data[i][col] > median:
print 1
else:
print 0
label = LABELS[col-16]
output_fp = os.path.join('data', 'matrix_data', 'logit', "wifi_features_" + label + '.csv')
fw = open(output_fp, 'a')
for i in range(data.shape[0]):
line = [str(x) for x in data[i,[3,7,8]].tolist()]
#print type(data[i][col])
if data[i][col] > median:
y = 1
else:
y = 0
line.append(str(y))
fw.write(','.join(line) + '\n')
fw.close()
if __name__ == '__main__':
fp = r"data\matrix_data\all_wifi_features.csv"
data = np.genfromtxt(fp, delimiter=",", dtype=float, skip_header=1)
x_cols = [1,2,3]
n = data.shape[0]
m = data.shape[1]
fold = 5
test_data = np.empty([n/fold, m])
train_data = np.empty([n-n/fold, m])
y_cols = range(4, 19)
for y_col in y_cols:
print '================================================================'
print LABELS[y_col-4]
replace_nan(data, y_col)
avg_mse = 0.0
for j in range(fold):
#print "fold %d" % j
for i, x in enumerate(data):
if i%fold == j:
test_data[i/fold] = data[i]
else:
train_data[(i/fold)*(fold-1)+i%fold-1] = data[i]
beta = get_beta(train_data, y_col)
mse = get_mse(test_data, beta, y_col)
avg_mse += mse
avg_mse /= fold
print "average mse: " + str(avg_mse)