def train_mnist_by_cnn():
# データの読み込み
(x_train, t_train), (x_test, t_test) = load_mnist(flatten=False)
# 処理に時間のかかる場合はデータを削減
x_train, t_train = x_train[:500], t_train[:500]
x_test, t_test = x_test[:100], t_test[:100]
max_epochs = 1
network = SimpleConvNet(input_dim=(1, 28, 28),
conv_param={
'filter_num': 30,
'filter_size': 5,
'padding': 0,
'stride': 1
},
hidden_size=100,
output_size=10,
weight_init_std=0.01)
trainer = Trainer(
network,
x_train,
t_train,
x_test,
t_test,
epoch_num=max_epochs,
batch_size=100,
optimizer='adagrad',
optimizer_param={'lr': 0.001},
evaluate_sample_num_per_epoch=None,
)
trainer.train()
def train_mnist_extend():
(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True)
# 過学習を再現するために、学習データを削減
x_train = x_train[:300]
t_train = t_train[:300]
input_size = 784
hidden_size_list = [100, 100, 100, 100, 100, 100]
output_size = 10
batch_size = 100
epoch_num = 301
network = MultiLayerNetExtend(input_size=input_size,
hidden_size_list=hidden_size_list,
output_size=output_size,
dropout=True,
dropout_ratio=0.45,
batch_normal=True)
logging.info(f'Layers: {network.layers.keys()}')
trainer = Trainer(
network=network,
x_train=x_train,
t_train=t_train,
x_test=x_test,
t_test=t_test,
epoch_num=epoch_num,
batch_size=batch_size,
optimizer="adagrad",
)
trainer.train()
logging.info(f'======== Train finished !! ========')
train_acc_list, test_acc_list = trainer.train_acc_list, trainer.test_acc_list
# グラフの描画==========
markers = {'train': 'o', 'test': 's'}
x = np.arange(len(train_acc_list))
plt.plot(x, train_acc_list, marker='o', label='train', markevery=10)
plt.plot(x, test_acc_list, marker='s', label='test', markevery=10)
plt.xlabel("epochs")
plt.ylabel("accuracy")
plt.xlim(0, epoch_num)
plt.ylim(0, 1.0)
plt.legend(loc='lower right')
plt.savefig("src/sample_data/mnist/params_backprop.png")
logging.info(f'======== Figure saved !! ========')
def train_mnist_two_layer():
(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True,
one_hot_label=True)
network = TwoLayerNetExtend(input_size=784, hidden_size=50, output_size=10)
iters_num = 10000
train_size = x_train.shape[0]
batch_size = 100
learning_rate = 0.1
train_loss_list = []
train_acc_list = []
test_acc_list = []
iter_per_epoch = max(train_size / batch_size, 1)
for i in range(iters_num):
batch_mask = np.random.choice(train_size, batch_size)
x_batch = x_train[batch_mask]
t_batch = t_train[batch_mask]
# 勾配
#grad = network.numerical_gradient(x_batch, t_batch)
grad = network.gradient(x_batch, t_batch)
# 更新
for key in ('W1', 'b1', 'W2', 'b2'):
network.params[key] -= learning_rate * grad[key]
loss = network.loss(x_batch, t_batch)
train_loss_list.append(loss)
if i % iter_per_epoch == 0:
train_acc = network.accuracy(x_train, t_train)
test_acc = network.accuracy(x_test, t_test)
train_acc_list.append(train_acc)
test_acc_list.append(test_acc)
print(train_acc, test_acc)
joblib.dump(network.params, "src/sample_data/mnist/params_backprop.pkl")
def load_data():
(x_train, t_train), (x_test, t_test) = mn.load_mnist(normalize=True,
one_hot_label=True)
return x_train, t_train, x_test, t_test
def train_cnn_by_pytorch():
network_path = "src/sample_data/mnist/convolution_network.pkl"
batch_size = 100
num_classes = 10
epochs = 3
(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True, one_hot_label=False)
network = ConvolutionalNeuralNetwork(num_classes)
# データのフォーマットを変換：PyTorchでの形式 = [画像数，チャネル数，高さ，幅]
x_train = x_train.reshape(60000, 1, 28, 28)
x_test = x_test.reshape(10000, 1, 28 ,28)
# PyTorchのテンソルに変換
x_train = torch.Tensor(x_train).float()
x_test = torch.Tensor(x_test).float()
t_train = torch.LongTensor(t_train) # labelにはint型(LongTensor型)を用いる ⇒ floatは×
t_test = torch.LongTensor(t_test)
# 学習用と評価用のデータセットを作成(60000枚分)
train_dataset = TensorDataset(x_train, t_train)
test_dataset = TensorDataset(x_test, t_test)
# データセットをバッチサイズ毎に分割する(ex: dataset:100, batchsize:20 ⇒ tensor(20 × 5)に分割)
train_batch = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
test_batch = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)
# パラメータ更新手法の定義
optimizer = op.Adagrad(network.parameters(), lr=0.01, lr_decay=0, weight_decay=0.05, initial_accumulator_value=0, eps=1e-10)
# 損失関数の定義
loss_func = nn.CrossEntropyLoss() # CrossEntropy誤差はone_hot_vectorに対応していない
if not os.path.exists(network_path):
network.train()
# 学習(エポック数3回 ⇒ パラメータは随時更新)
for i in range(1, epochs+1):
logging.info(f'===== START {i}th epoch !! =====')
for i, (data, label) in enumerate(train_batch):
optimizer.zero_grad()
output = network(data)
loss = loss_func(output, label)
loss.backward()
optimizer.step()
if i % 100 == 0:
logging.info(f'{i}th iteration ⇒ loss: {loss.item()}')
joblib.dump(network, network_path)
else:
network = joblib.load(network_path)
# 評価
network.eval() # モデルを推論モードに変更
count = 0
for i, (data, label) in enumerate(test_batch):
test_output = network(data)
_, predicted = torch.max(test_output.data, 1)
y_predicted = predicted.numpy() # tensor型 ⇒ numpy.array型
label = label.numpy() # tensor型 ⇒ numpy.array型
cnt = np.sum(y_predicted == label)
count += cnt
accuracy = count / len(t_test) * 100
logging.info(f'========= Finaly accuracy is {accuracy} !! =========')