def main():
# Number of possible degrees to be tested
K = 30
data_path = 'data_linreg.json'
# Load the data
f = open(data_path, 'r')
data = json.load(f)
for k, v in data.items():
data[k] = np.array(v).reshape((len(v), 1))
# Init vectors storing MSE (Mean square error) values of each sets at each degrees
mse_train = np.zeros(K)
mse_val = np.zeros(K)
mse_test = np.zeros(K)
theta_list = np.zeros(K, dtype=object)
degrees = np.arange(K) + 1
# Compute the MSE values
for i in range(K):
theta_list[i], mse_train[i], mse_val[i], mse_test[
i] = poly.train_and_test(data, degrees[i])
i_best = np.argmin(mse_val)
best_degree = degrees[i_best]
best_theta = theta_list[i_best]
# Plot the training error as a function of the degrees
plot_errors(i_best, degrees, mse_train, mse_val, mse_test)
plot_poly(data, best_degree, best_theta)
plt.show()
def main():
# Number of possible degrees to be tested
K = 30
data_path = 'data_linreg.json'
# Load the data
f = open(data_path, 'r')
data = json.load(f)
for k, v in data.items():
data[k] = np.array(v).reshape((len(v), 1))
# Init vectors storing MSE (Mean squared error) values of each set at each degrees
mse_train = np.zeros(K)
mse_val = np.zeros(K)
mse_test = np.zeros(K)
theta_list = np.zeros(K, dtype=object)
degrees = np.arange(K) + 1
# Compute the MSE values
for i in range(K):
theta_list[i], mse_train[i], mse_val[i], mse_test[i] = poly.train_and_test(data, degrees[i])
######################
#
# TODO
#
# Find the best degree that minimizes the validation error.
# Store it in the variable i_best for plotting the results
#
# TIPs:
# - use the argmin function of numpy
# - the code above is already giving the vectors of errors
i_best_val = np.argmin(mse_val) # TODO: Change this
best_degree_val = degrees[i_best_val]
best_theta_val = theta_list[i_best_val]
i_best_tr = np.argmin(mse_train) # TODO: Change this
best_degree_tr = degrees[i_best_tr]
best_theta_tr = theta_list[i_best_tr]
#
# END TODO
######################
# Plot the training error as a function of the degrees
plot_errors(i_best_val, degrees, mse_train, mse_val, mse_test)
plot_poly(data, best_degree_val, best_theta_val)
plt.show()
plot_errors(i_best_tr, degrees, mse_train, mse_val, mse_test)
plot_poly(data, best_degree_tr, best_theta_tr)
plt.show()
def main():
# Set the degree of the polynomial expansion
degree = 1
degrees = []
for j in range(30):
degrees.append(j)
for i in range(1, 31):
degree = i
data_path = 'data_linreg.json'
# Load the data
f = open(data_path, 'r')
data = json.load(f)
for k, v in data.items():
data[k] = np.array(v).reshape((len(v), 1))
# Print the training and testing errors
theta, err_train, err_val, err_test = poly.train_and_test(data, degree)
print('Training error {} \t Validation error {} \t Testing error {} '.
format(err_train, err_val, err_test))
if (degree == 1):
lowest = err_train
if (lowest > err_train):
lowest = err_train
lowest_degree = degree
lowest_theta = theta
#def plot_poly(data, degree, theta_opt, n_line_precision=100):
#def plot_errors(i_best, degrees, mse_train, mse_val, mse_test):
plot_errors(13, degrees, err_train, err_val, err_test)
plt.show()