[1, 0, 1],
[1, 1, 1]])
Y = np.array([0, 0, 1, 1])
nn = NN(3, [df_train.shape[1], 50, 1], 3 * [sigma], 3 * [dsigma])
X = df_train
Y = df["spam_or_not"]
errors = []
result_errors = []
for i in range(1000, 50000):
k = np.random.randint(0, 4600)
nn.train(X[k], Y[k], 1000. / i)
if i % 10000 == 0:
error = nn.error(X[:1000], Y[:1000])
errors.append(error)
print("Epoch: {}; Error: {}".format(i, error))
result_errors.append(nn.result_error(X[:1000], Y[:1000]))
with open('nn.pkl', 'wb') as f:
dill.dump(nn,f)
print(nn.error(X, Y))
plt.plot(errors)
plt.plot(result_errors, "r")
plt.ylim([0, 1])
plt.show()
errors_01 = []
errors_001 = []
errors_1N = []
#############
# eta = 0.1 #
#############
nn = NN(4, [df_train.shape[1], 50, 50, 1], 4 * [sigma], 4 * [dsigma])
for i in range(5000, 1000000):
k = np.random.randint(0, len(X_train))
nn.train(X_train[k], Y_train[k], 0.1)
if i % 10000 == 0:
error = nn.error(X_test, Y_test)
errors_01.append(error)
print("Epoch: {}; Error: {}".format(i, error))
##############
# eta = 0.01 #
##############
nn = NN(4, [df_train.shape[1], 50, 50, 1], 4 * [sigma], 4 * [dsigma])
for i in range(5000, 1000000):
k = np.random.randint(0, len(X_train))
nn.train(X_train[k], Y_train[k], 0.01)
if i % 10000 == 0:
error = nn.error(X_test, Y_test)
[0, 1, 1],
[1, 0, 1],
[1, 1, 1]])
Y = np.array([0, 0, 1, 1])
for eta in [1, 0.5, 0.1, 0.05, 0.01, 0.005, 0.001]:
errors = []
nn = NN(2, [3, 1], 2 * [sigma], 2 * [dsigma])
for i in range(10000):
k = np.random.randint(0, 4)
nn.train(X[k], Y[k], eta)
if i % 10 == 0:
error = nn.error(X, Y)
errors.append(error)
print("Epoch: {}; Error: {}".format(i, error))
plt.plot([10 * i for i in range(len(errors))], errors, label = "eta = {}".format(eta))
plt.legend()
plt.xlabel("Epoch")
plt.ylabel("Error")
plt.title("Basic")
plt.ylim([0, 1])
plt.show()
X = np.array([[0, 0, 1],
[0, 1, 1],
[1, 0, 1],
[1, 1, 1]])
