Question

我正在使用auto-mpg.data。我遇到了这个问题：

回溯（最近通话最近）：在第60行的文件“ C：/PythonProjects/test5.py” _，c = sess.run（[优化器，损失]，feed_dict = {X：xTrain，Y：yTrain}）＃获取损失值运行中的文件“ C：\ Python3 \ lib \ site-packages \ tensorflow \ python \ client \ session.py”，第929行 run_metadata_ptr） _run中的文件“ C：\ Python3 \ lib \ site-packages \ tensorflow \ python \ client \ session.py”，行1128 str（subfeed_t.get_shape（））））

ValueError：无法为形状为（（,,）'的张量'Placeholder_1：0'输入形状（352，1）的值

我尝试为Xtrain，Xtest，Ytrain，Ytest创建单独的占位符，但是我认为这也不是正确的方法。

如何为测试/训练数据使用相同的X和Y占位符？

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split


filename = "auto-mpg.data"
column_names = ['mpg', 'cylinders', 'displacement', 'horsepower', 'weight', 'acceleration', 'year', 'origin', 'name']
df = pd.read_csv(filename, delim_whitespace=True, header=None, na_values = "?", names=column_names)
df = df.drop('name', axis=1)
df = df.dropna()

# Ont-hot encoding for category data
origin = df.pop('origin')
df['USA'] = (origin == 1)*1.0
df['Europe'] = (origin == 2)*1.0
df['Japan'] = (origin == 3)*1.0


df = df.drop(['year'], axis=1)
x_data = df.drop('mpg', axis=1)
y_data = df[['mpg']] # Continuous target variable : mpg

print(df.shape)


# Test/Train split of 90%/10%
xTrain, xTest, yTrain, yTest = train_test_split(x_data, y_data, test_size=0.1, random_state=0)

print(xTrain.shape)  # 352x8
print(xTest.shape)  # 40x8
print(yTrain.shape)  # 352x1
print(yTest.shape)  #40x1

def LinearRegression():
    y_pred = tf.add(tf.matmul(X, W), b)
    loss = tf.reduce_mean(tf.square(y_pred - Y))
    return loss

# Xtrain = tf.placeholder(tf.float32, [352,8])
# Ytrain = tf.placeholder(tf.float32, [352,1])
# Xtest = tf.placeholder(tf.float32, [40,8])
# Ytest = tf.placeholder(tf.float32, [40,1])
numFeatures = xTrain.shape[1]
X = tf.placeholder(tf.float32, [None,numFeatures])
Y = tf.placeholder(tf.float32, [None])

W = tf.get_variable(name='Weight', dtype=tf.float32, shape=([8,1]), initializer=tf.zeros_initializer())
b = tf.get_variable(name='Bias', dtype=tf.float32, shape=([1]), initializer=tf.zeros_initializer())
loss=LinearRegression()
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.0000001).minimize(loss)
epochs = 1000
display_step = 100  # Display every 10s output

with tf.Session() as sess:
    init = tf.global_variables_initializer()
    sess.run(init)
    for e in range(epochs):
        _, c = sess.run([optimizer, loss], feed_dict={X: xTrain, Y: yTrain})  # Get loss value
        if (e + 1) % display_step == 0:
            print('Epoch #:', '%d' % (e + 1), 'Loss =', '{:.9f}'.format(c), 'W =', sess.run(W), 'b =', sess.run(b))

    print("Training completed...")

    training_cost = sess.run(loss, feed_dict={X: xTrain, Y: yTrain})
    weight = sess.run(W)
    bias = sess.run(b)
    print("Training cost=", training_cost, '; ' "W =", weight, '; ' "b =", bias)

    print("Testing result...")
    test_loss = LinearRegression()  # Same function as above
    testing_cost = sess.run(test_loss, feed_dict={X: xTest, Y: yTest})
    print("Testing cost:", testing_cost)
    print("Absolute mean square loss difference:", abs(training_cost - testing_cost))

    fitted_prediction = sess.run(W) * xTest + sess.run(b)
    print('fitted_prediction = ',fitted_prediction)

Answer 1

您的yTrain似乎排名2（[352,1]），但您尝试与其一起使用的占位符是标量。尝试将Y更改为

Y = tf.placeholder(tf.float32, [None,1])

TensorFlow多元线性回归占位符和矩阵乘法误差

1 个答案: