张量重量直方图与正态分布不同

Question

我无法理解为什么权重不像正态分布。

实际上我想了解体重变化过程中发生了什么，以及渐变发生了什么。但问题是直方图中的权重看起来不像正态分布。

你可以找到代码：

iris_data_set = pd.read_csv('iris.csv')
iris_data_set.head()
cols_to_norm = ['Sepal.Length' , 'Sepal.Width' , 'Petal.Length' , 
'Petal.Width']
iris_data_set[cols_to_norm] = iris_data_set[cols_to_norm].apply(lambda x:(x- 
 x.min()) / (x.max() - x.min()))
feat_data = iris_data_set.drop('Species', axis=1 )
label = iris_data_set['Species']
X_train, X_test, y_train, y_test = train_test_split(feat_data , label, 
test_size = 0.3 , random_state =101)

y_train = pd.get_dummies(y_train)
y_test = pd.get_dummies(y_test)

n_features = 4 
n_dense_neurons = 3
n_output = 3
training_steps =1000
#tf Graph input
X_data = tf.placeholder(tf.float32 , shape= [None , n_features], 
name='Inputdata')
y_target = tf.placeholder(tf.float32 , shape= [None , n_output], 
name='Labeldata')
#Store layers

weights = {
'w1':  tf.Variable(tf.random_normal(shape=[n_features , n_dense_neurons]) , 
name = 'w1'), # Inputs -> Hidden Layer
'w2':  tf.Variable(tf.random_normal(shape=[n_dense_neurons , n_output]) , 
name = 'w2')
}

biases = {       
   'b1': tf.Variable(tf.random_normal(shape=[n_dense_neurons]) , 
name='b1'),   # First Bias
   'b2': tf.Variable(tf.random_normal(shape=[n_output]) , name='b2')
}

def multilayer_perceptron(x, weights, biases):
   # Hidden layer with RELU activation
    layer_1 = tf.add(tf.matmul(X_data , weights['w1']), biases['b1'])
   layer_1 = tf.nn.relu(layer_1)
   # Create a summary to visualize the first layer ReLU activation
   tf.summary.histogram("relu1", layer_1)

# Output layer
out_layer = tf.add(tf.matmul(layer_1, weights['w2']), biases['b2'])
return out_layer
with tf.name_scope('Model'):
   pred = multilayer_perceptron(X_data, weights, biases)
with tf.name_scope('Loss'):
   final_output = tf.nn.softmax(pred)
   deltas = tf.square (final_output - y_target)
   loss = tf.reduce_sum (deltas)
with tf.name_scope('SGD'):
    optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.001)
    grads = tf.gradients(loss, tf.trainable_variables())
    grads = list(zip(grads, tf.trainable_variables()))
    apply_grads = optimizer.apply_gradients(grads_and_vars=grads)
with tf.name_scope('Accuracy'):
     acc = tf.equal(tf.argmax(pred, 1), tf.argmax(y_target, 1))
     acc = tf.reduce_mean(tf.cast(acc, tf.float32))
init = tf.global_variables_initializer()

tf.summary.scalar("loss", loss)

tf.summary.scalar("accuracy", acc)

for var in tf.trainable_variables():
    tf.summary.histogram(var.name, var)
for grad, var in grads:
    tf.summary.histogram(var.name + '/gradient', grad)
 # Merge all summaries into a single op
merged_summary_op = tf.summary.merge_all()
with tf.Session() as sess: 
   sess.run(init)
   summary_writer= tf.summary.FileWriter("new6", 
   for i in range (training_steps):graph=tf.get_default_graph())
   _, c, summary = sess.run([apply_grads, loss, merged_summary_op],
                                feed_dict={X_data: X_train, y_target: 
   y_train})


if i % 20 == 0:
    summary_str = sess.run(merged_summary_op, feed_dict={X_data: 
   X_train, y_target: y_train})
    summary_writer.add_summary(summary_str, i)
    summary_writer.flush()