tensorflowの最適化計算時に渡すデータについて

データはbostonデータセットを用います。

Python
1from sklearn.datasets import load_boston
2from sklearn.model_selection import train_test_split
3boston = load_boston()
4X_train, X_test, y_train, y_test = X_train, X_test, y_train, y_test = train_test_split(boston['data'], boston['target'], test_size=0.3,  random_state=0)
5
6X_train = X_train.astype(np.float32)
7X_test = X_test.astype(np.float32)
8y_train = y_train.astype(np.float32)
9y_test = y_test.astype(np.float32)

最適化を下記で実行します。

Python
1# 回帰係数の定義
2W = tf.Variable(tf.random_uniform([1], -1.0, 1.0))
3# 定数項の定義
4b = tf.Variable(tf.zeros([1]))
5# 回帰式の定義
6y = W * X_train[:,0] + b
7
8# 損失関数の定義
9loss = tf.reduce_mean(tf.square(y - y_train))
10# 最適化基準の定義
11optimizer = tf.train.GradientDescentOptimizer(0.5)
12train = optimizer.minimize(loss)
13
14with tf.Session() as sess:
15    # 変数を使う場合は初期化が必要
16    sess.run(tf.global_variables_initializer())
17    # 学習開始
18    for step in range(201):
19        W_new = sess.run(W)
20        b_new = sess.run(b)
21        sess.run(train)
22        if step % 20 == 0:
23            print(step, sess.run(W), sess.run(b))

出力は

Python
10 [92.415306] [24.678692]
220 [inf] [inf]
340 [nan] [nan]
460 [nan] [nan]
580 [nan] [nan]
6100 [nan] [nan]
7120 [nan] [nan]
8140 [nan] [nan]
9160 [nan] [nan]
10180 [nan] [nan]
11200 [nan] [nan]

となります。
しかし、下記の様にデータを生成してやると、うまくいきます。

Python
1X_train = X_train.astype(np.float32)/100
2X_test = X_test.astype(np.float32)/100
3y_train = y_train.astype(np.float32)/100
4y_test = y_test.astype(np.float32)/100