def evaluate(X_train, X_test, y_train, y_test, alpha):
# Implement your method here
model = Lasso(alpha=alpha)
model.fit(X_train,y_train)
MSEsum = 0
y_pred = model.predict(X_test)
for i in range(len(y_test)):
MSEsum+=(y_pred[i]-y_test[i])**2
return MSEsum/len(y_test)
def deploy_lasso(sample_id):
sample_data = sample_data_list[sample_id]
sample_x, sample_y = sample_data.x, sample_data.y
sample_phi = generate_polynomial_features(sample_x)
title = 'LASSO'
log = open(result_sub_path + title + '_' + str(sample_id) + '.txt', 'w')
img_path = result_sub_path + title + '_' + str(sample_id) + '.png'
candidate_pred_y, candidate_error, candidate_sample_error, candidate_lambda = [], [], [], []
_lambda = 0.000001
_lambda_step = 1.05
log.write("range of lambda: " + str(_lambda) + ' : ')
for candidate_id in range(1000):
model = Lasso()
model.fit(sample_phi, sample_y, _lambda)
pred_sample_y = model.predict(sample_phi)
pred_y = model.predict(gt_phi)
sample_error = compute_mean_squared_error(pred_sample_y, sample_y)
error = compute_mean_squared_error(pred_y, gt_y)
candidate_sample_error.append((sample_error, candidate_id))
candidate_error.append(error)
candidate_lambda.append(_lambda)
candidate_pred_y.append(pred_y)
_lambda *= _lambda_step
log.write(str(_lambda) + '\n')
candidate_error.sort()
(sample_error, best_id) = candidate_sample_error[0]
error = candidate_error[best_id]
pred_y = candidate_pred_y[best_id]
_lambda = candidate_lambda[best_id]
log.write("lambda: " + str(_lambda) + '\n')
log.write("Training MSE: " + str(sample_error) + '\n')
log.write("Testing MSE: " + str(error) + '\n')
visualize(sample_x, sample_y, gt_x, gt_y, pred_y, title, img_path)
return sample_error, error
def deploy_lasso():
title = 'LASSO'
log = open(result_sub_path + title + '_2.txt', 'w')
img_path = result_sub_path + title + '_2.png'
candidate_pred_y, candidate_error, candidate_sample_error, candidate_lambda = [], [], [], []
_lambda = 1.0
_lambda_step = 1.05
log.write("range of lambda: " + str(_lambda) + ' : ')
for candidate_id in range(10):
model = Lasso()
model.fit(sample_phi, sample_y, _lambda)
pred_sample_y = model.predict(sample_phi)
pred_y = model.predict(gt_phi)
sample_error = compute_mean_squared_error(pred_sample_y, sample_y)
error = compute_mean_squared_error(pred_y, gt_y)
candidate_sample_error.append((sample_error, candidate_id))
candidate_error.append(error)
candidate_lambda.append(_lambda)
candidate_pred_y.append(pred_y)
_lambda *= _lambda_step
log.write(str(_lambda) + '\n')
candidate_error.sort()
(sample_error, best_id) = candidate_sample_error[0]
error = candidate_error[best_id]
pred_y = candidate_pred_y[best_id]
_lambda = candidate_lambda[best_id]
log.write("lambda: " + str(_lambda) + '\n')
log.write("Training MSE: " + str(sample_error) + '\n')
log.write("Testing MSE: " + str(error) + '\n')
sample_error = compute_mean_absolute_error(pred_sample_y, sample_y)
error = compute_mean_absolute_error(pred_y, gt_y)
log.write("Training MAE: " + str(sample_error) + '\n')
log.write("Testing MAE: " + str(error) + '\n')
visualize_people_counting(gt_y, pred_y, title, img_path)
#! -*- coding: utf-8 -*-
import pandas as pd
import numpy as np
from sklearn import cross_validation
from lasso import Lasso
df = pd.read_csv("Boston.csv", index_col=0)
y = df.iloc[:, 13].values
df = (df - df.mean()) / df.std() # 基準化
X = df.iloc[:, :13].values
model = Lasso(alpha=1.0, max_iter=1000)
model.fit(X, y)
print(model.intercept_)
print(model.coef_)
scores = cross_validation.cross_val_score(model,
X,
y,
cv=5,
scoring='mean_squared_error')
print(np.mean(scores))
#! -*- coding: utf-8 -*-
import pandas as pd
import numpy as np
from sklearn import cross_validation
from lasso import Lasso
df = pd.read_csv("Boston.csv", index_col=0)
y = df.iloc[:, 13].values
df = (df - df.mean())/df.std() # 基準化
X = df.iloc[:, :13].values
model = Lasso(alpha=1.0, max_iter=1000)
model.fit(X, y)
print model.intercept_
print model.coef_
scores = cross_validation.cross_val_score(model, X, y, cv=5, scoring='mean_squared_error')
print np.mean(scores)
