def test_accuracysgd(idx):
test_X, test_y = readmnist.testadata()
test_X = construct_matrix(test_X)
test_y = construct_labels(test_y)
# test_X = normalizedata(test_X)
def test_model(network):
test_predictions = network.run(test_X)
correct = 0
total = 0
for i in range(len(test_predictions)):
total += 1
if np.argmax(np.array(test_predictions[i])) == test_y[i]:
correct += 1
return correct/total
print(test_model(network_sgd))
def accuracy_train(w):
data_test, label_test = readmnist.testadata()
test = construct_matrix(data_test)
x = np.insert(test, 0, 1, axis=1)
label = construct_labels(label_test)
limit = test.shape[0]
counter = 0
for i in range(limit):
score = np.zeros((10))
for j in range(10):
score[j] = np.matmul(x[i], w[:, j])
lb_pred = np.argmax(score)
if lb_pred == label[i]:
counter += 1
accuracy = counter / limit
return accuracy
t_corr_labels = []
for i in range(te_size):
if t_lbls[i] == test_y[i]:
t_corr_labels.append(1)
t_acc = 1.0 * np.array(t_corr_labels).sum() / te_size
print(t_acc)
if __name__ == "__main__":
train_X, train_y = readdata()
idx = np.random.randint(train_X.shape[0], size=60000)
train_X = train_X[idx, :]
train_y = train_y[idx, :]
data_test, label_test = readmnist.testadata()
test = construct_matrix(data_test)
label_test = construct_labels(label_test)
nonlinear_svm(train_X, train_y, data_test, label_test,
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
train_X, train_y = readdata()
idx = np.random.randint(train_X.shape[0], size=60000)
train_X = train_X[idx, :]
train_y = train_y[idx, :]
data_test, label_test = readmnist.testadata()
test = construct_matrix(data_test)
label_test = construct_labels(label_test)
print(accuracy_train(multiclass(10, test, label_test), train_X, train_y))