epochs = 10
for epoch in range(epochs):
time_start = time.time()

for images, labels in dataset: # 1step分のデータを取り出し         train_step((images, labels)) # 1step分の学習を実行
# 平均ロスと平均精度     # Metric.result()メソッドで取り出せる     epoch_loss = train_loss.result()
epoch_acc = 100 * train_acc.result()

# epochの結果を表示     time_epoch = time.time() – time_start
print(‘epoch: {:} loss: {:.4f} acc: {:.2f}% time: {:.2f}s’.format(
epoch + 1, epoch_loss, epoch_acc, time_epoch))

# epoch毎にリセットしないと累積していく     train_loss.reset_states()
train_acc.reset_states()

# 学習済みの重みを保存 model.save_weights(‘weights_{:}’.format(epochs))

#========================================================================= # numpy配列を直接与えて学習を実行 #=========================================================================
print(‘train with numpy array.’)

epochs = 10
for epoch in range(epochs):
time_start = time.time()

# ここから違う========================================================
indices = np.arange(N)
np.random.shuffle(indices)
for i in range(N // batch_size):
# 1step分のデータを取り出し         indices_batch = indices[i * batch_size🙁i + 1) * batch_size]
images = X_train[indices_batch]
labels = Y_train[indices_batch]

# 1step分の学習を実行         # numpy配列を直接入れてOK         train_step((images, labels))

# ここまで違う========================================================
# 平均ロスと平均精度     # Metric.result()メソッドで取り出せる     epoch_loss = train_loss.result()
epoch_acc = 100 * train_acc.result()

# epochの結果を表示     time_epoch = time.time() – time_start
print(‘epoch: {:} loss: {:.4f} acc: {:.2f}% time: {:.2f}s’.format(
epoch + 1, epoch_loss, epoch_acc, time_epoch))

# epoch毎にリセットしないと累積していく     train_loss.reset_states()
train_acc.reset_states()

# 学習済みの重みを保存 model.save_weights(‘weights_{:}’.format(epochs))

# グラフの作成
x = tf.compat.v1.placeholder(tf.float32, [None, 784]) # 入力するPlaceholder tf.placeholder
W = tf.Variable(tf.zeros([784, 10])) # 重み
b = tf.Variable(tf.zeros([10])) # バイアス
y = tf.matmul(x, W) + b # 内積計算とバイアスの加算

# 正解ラベル用のplaceholeder
y_ = tf.compat.v1.placeholder(tf.float32, [None, 10]) # tf.placeholder

# 損失の計算方法と、オプティマイザーを定義
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y))
train_step = tf.compat.v1.train.GradientDescentOptimizer(0.5).minimize(cross_entropy) # tf.train.GradientDescentOptimizer

# セッションの作成と初期化
sess = tf.compat.v1.InteractiveSession() # tf.InteractiveSession()
tf.compat.v1.global_variables_initializer().run() # tf.global_variables_initializer().run()

# 学習部(1000回学習)
for _ in range(1000):
batch_xs, batch_ys = mnist.train.next_batch(100)
sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})

# テスト部
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print(sess.run(accuracy, feed_dict={x: mnist.test.images,
y_: mnist.test.labels}))

Future major versions of TensorFlow will allow gradients to flow
into the labels input on backprop by default.

See `tf.nn.softmax_cross_entropy_with_logits_v2`.

2019-12-29 17:40:40.072583: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
0.9189

Process finished with exit code 0