1）偏差形状

Question

有两件事我无法理解。以下代码应该学习简单的OR逻辑功能：

import numpy as np
import tensorflow as tf
from datetime import datetime

now = datetime.utcnow().strftime("%Y%m%d%H%M%S")
root_logdir = "tf_logs"
logdir = "{}/run-{}/".format(root_logdir, now)

def neuron_layer(X, n_neurons, name, activation = None):
    with tf.name_scope(name):
        n_inputs = int(X.get_shape()[1])
        stddev = 2 / np.sqrt(n_inputs)
        init = tf.truncated_normal((n_inputs,n_neurons), stddev= stddev)
        W = tf.Variable(init, name="kernel")
        b = tf.Variable(tf.zeros([n_neurons]), name="bias")
        Z = tf.matmul(X,W) + b
        if activation is not None:
            return activation(Z)
        else:
            return Z

n_INPUTS = 2
X_OR = np.array([[0,0],[0,1],[1,0],[1,1]])
y_OR = np.array([[0], [1], [1],[1]])

X_train = X_OR
y_train = y_OR

with tf.name_scope('input'):
X = tf.placeholder(tf.float32, [None, n_INPUTS], name="X-inputs")
y = tf.placeholder(tf.float32, [None,1], name = "y-inputs")


# hidden1 = neuron_layer(X, 2,"hidden1")
y_pred = neuron_layer(X,1,"output")

with tf.name_scope('MSE'):
    error = y_pred - y
    mse = tf.reduce_mean(tf.square(error), name="mse")
    tf.summary.scalar('mse-scalar', mse)

with tf.name_scope('train'):
    optimiser = tf.train.AdamOptimizer(0.01)
    training_op = optimiser.minimize(mse)

merged = tf.summary.merge_all()
train_writer = tf.summary.FileWriter(logdir + '/train',             
tf.get_default_graph())
test_writer = tf.summary.FileWriter(logdir + '/test')

init = tf.global_variables_initializer()

n_epochs = 1000

with tf.Session() as sess:
    sess.run(init)

    for epoch in range(n_epochs):
        summary, _ = sess.run([merged, training_op],
                              feed_dict={X:X_train, y: y_train})
        train_writer.add_summary(summary, epoch)
        if epoch % 100 == 99:
            y_pred_test = y_pred.eval(feed_dict={X: X_train, y: y_train})
            print(y_pred_test)


train_writer.close()
test_writer.close()

1）偏差形状

目前，我的偏差形状遵循以下规则：偏差节点的数量等于层中神经元的数量。因此b = tf.Variable(tf.zeros([n_neurons]), name="bias") 但是，在这种情况下，网络无法收敛。得到的结果等于：

[[0.2599482 ]
 [0.7518728 ]
 [0.75169754]
 [1.2436221 ]]

但是，如果我将其形状更改为b = tf.Variable(tf.zeros([4, n_neurons]), name="bias")（其中4是数据点数），那么它将收敛。但是，据我所记得，等式为X * w + b，其中X的形状为（示例数，参数数量），w的形状为（参数数，神经元数量），b的形状为（数神经元）。我还可以从文档（https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/tutorials/mnist/mnist.py）中注意到该示例。 为什么我需要批量指定偏差形状的示例数量？我不应该这样。

2）Y占位符形状。

给出链接中的MNIST示例，我希望y占位符具有以下形状：

 y = tf.placeholder(tf.float32, [None], name = "y-inputs")

因为我们通常不知道许多示例，所以目标只是类的列表。但是，如果在我的代码中进行更改，我将得到两个不同的结果：

（正确的结果）

X_OR = np.array([[0,0],[0,1],[1,0],[1,1]])
y_OR = np.array([[0], [1], [1],[1]])

X = tf.placeholder(tf.float32, [None, n_INPUTS], name="X-inputs")
y = tf.placeholder(tf.float32, [None,1], name = "y-inputs")

到（错误结果）：

X_OR = np.array([[0,0],[0,1],[1,0],[1,1]])
y_OR = np.array([0, 1, 1,1])

X = tf.placeholder(tf.float32, [None, n_INPUTS], name="X-inputs")
y = tf.placeholder(tf.float32, [None], name = "y-inputs")

tf如何解释这种形状变化？

Tensorflow-占位符尺寸和偏置形状

1）偏差形状

2）Y占位符形状。

0 个答案: