tensorflow NNの評価用データの出力がnan

Question

以下コードで実行したところ、評価用(x_vals_test, y_vals_test)が上手く働いてくれません。
x_vals_test, y_vals_testの損失がnanと出力されるのはなぜなのでしょうか。
分かる方がいれば教えて頂けると幸いです。

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

x_vals_train, x_vals_test, y_vals_train, y_vals_test = train_test_split(df.as_matrix(), df_Score.as_matrix(), test_size=0.2, random_state=0)

# Record training column max and min for scaling of non-training data
train_max = np.max(x_vals_train, axis=0)
train_min = np.min(x_vals_train, axis=0)

# Normalize by column (min-max norm to be between 0 and 1)
def normalize_cols(mat, max_vals, min_vals):
    return (mat - min_vals) / (max_vals - min_vals)

x_vals_train = np.nan_to_num(normalize_cols(x_vals_train, train_max, train_min))
x_vals_test = np.nan_to_num(normalize_cols(x_vals_test, train_max, train_min))

# Define Variable Functions (weights and bias)
def init_weight(shape, st_dev):
    weight = tf.Variable(tf.random_normal(shape, stddev=st_dev))
    return(weight)
    
def init_bias(shape, st_dev):
    bias = tf.Variable(tf.random_normal(shape, stddev=st_dev))
    return(bias)
    
# Create Placeholders
x_data = tf.placeholder(shape=[None, 3806], dtype=tf.float32)
y_target = tf.placeholder(shape=[None, 1], dtype=tf.float32)

# Create a fully connected layer:
def fully_connected(input_layer, weights, biases):
    layer = tf.add(tf.matmul(input_layer, weights), biases)
    return(tf.nn.relu(layer))

def final_connected(input_layer, weights, biases):
    layer = tf.add(tf.matmul(input_layer, weights), biases)
    return layer
  
#--------Create the first layer (50 hidden nodes)--------
weight_1 = init_weight(shape=[3806, 10], st_dev=10.0)
bias_1 = init_bias(shape=[10], st_dev=10.0)
layer_1 = fully_connected(x_data, weight_1, bias_1)

#--------Create second layer (25 hidden nodes)--------
weight_2 = init_weight(shape=[10, 5], st_dev=10.0)
bias_2 = init_bias(shape=[5], st_dev=10.0)
layer_2 = fully_connected(layer_1, weight_2, bias_2)

#--------Create third layer (5 hidden nodes)--------
weight_3 = init_weight(shape=[5, 3], st_dev=10.0)
bias_3 = init_bias(shape=[3], st_dev=10.0)
layer_3 = fully_connected(layer_2, weight_3, bias_3)

#--------Create output layer (1 output value)--------
weight_4 = init_weight(shape=[3, 1], st_dev=10.0)
bias_4 = init_bias(shape=[1], st_dev=10.0)
final_output = final_connected(layer_3, weight_4, bias_4)

# Declare loss function (L1)
loss = tf.reduce_mean(tf.abs(y_target - final_output))

# Declare optimizer
my_opt = tf.train.AdamOptimizer(0.001)
train_step = my_opt.minimize(loss)

# Initialize Variables
init = tf.global_variables_initializer()
sess.run(init)

# Training loop
loss_vec = []
test_loss = []
for i in range(80000):
    rand_index = np.random.choice(len(x_vals_train), size=batch_size)
    rand_x = x_vals_train[rand_index]
    rand_y = np.transpose([y_vals_train[rand_index]])
    sess.run(train_step, feed_dict={x_data: rand_x, y_target: rand_y})

    temp_loss = sess.run(loss, feed_dict={x_data: rand_x, y_target: rand_y})
    loss_vec.append(temp_loss)
   
    test_temp_loss = sess.run(loss, feed_dict={x_data: x_vals_test, y_target: np.transpose([y_vals_test])})
    test_loss.append(test_temp_loss)
    
    if (i+1) % 20000 == 0:
        print('Generation: ' + str(i+1) + '. Loss = ' + str(temp_loss) + '. TEST_LOSS ' + str(test_temp_loss))

Generation: 20000. Loss = 0.6681944. TEST_LOSS nan
Generation: 40000. Loss = 0.6850753. TEST_LOSS nan
Generation: 60000. Loss = 0.71310586. TEST_LOSS nan
Generation: 80000. Loss = 0.82346433. TEST_LOSS nan