Styjun

Draw a symmetric alien head

What preparations would Hubble have needed to return in a Shuttle?

In a list with unique pairs A, B, how can I sort them so that the last B is the first A in the next pair?

Leaving job close to major deadlines

How "fast" do astronomical events occur?

Why there is a red color in right side?

What is the highest power supply a Raspberry pi 3 B can handle without getting damaged?

Setting up the trap

In the US, can a former president run again?

I just entered the USA without passport control at Atlanta airport

Scaling an object to change its key

What kind of chart is this?

How did Frodo know where the Bree village was?

Bent arrow under a node

How to compute the inverse of an operation in Q#?

In Street Fighter, what does the M stand for in M Bison?

reverse a call to mmap()

How to write a nice frame challenge?

Unrecognized IC Package Style

How can I prevent a user from copying files on another hard drive?

First occurrence in the Sixers sequence

Is Newton's third law really correct?

How to make all magic-casting innate, but still rare?

Why is it 出差去 and not 去出差?

weights initialization in tensorflow for n-dimensional input

Use tensorflow tf.Variable instead of tf.placeholder for train dataTensorflow: Input to reshape is a tensor with 79744 values, but the requested shape has 203392How to reshape the pre-trained weights to input them to 3d convoluional neural network?Simple Feedforward Neural Network with TensorFlow won't learnCan't run prediciton because of troubles with tf.placeholderpython tensorflow: How to avoid giving feed_dict to every sess.run()?Storing TensorFlow network weights in Python multi-dimensional listsTensorFlow object detection api: classification weights initialization when changing number of classes at training using pre-trained modelsHow to implement pre-training in Tensorflow? How to partially use saved weights from checkpoint file?Can you Transpose/Reverse the shape in Tensorflow's placeholder?

.everyoneloves__top-leaderboard:empty,.everyoneloves__mid-leaderboard:empty,.everyoneloves__bot-mid-leaderboard:empty height:90px;width:728px;box-sizing:border-box;

0

training set x_train contain 10k examples and each input is of shape(10,23201)
and when i try to send the total training set for training it giving an error
but when i send 10 examples at a time it working fine.
How to change the code that will take all examples at once(what shape of weights will change my probllem)?

input_features = 23201
hidden_layer_1 = 300
hidden_layer_2 = 300
hidden_layer_3 = 128

limit_1 = tf.cast(np.sqrt(6/(input_features+hidden_layer_1)), tf.float32)
limit_2 = tf.cast(np.sqrt(6/(hidden_layer_1+hidden_layer_2)), tf.float32)
limit_3 = tf.cast(np.sqrt(6/(hidden_layer_2+hidden_layer_3)), tf.float32)
#weights initialization
w0 = tf.Variable(tf.random_uniform([10, input_features, hidden_layer_1], -limit_1, limit_1))
w1 = tf.Variable(tf.random_uniform([10, hidden_layer_1, hidden_layer_2],-limit_2, limit_2))
w2 = tf.Variable(tf.random_uniform([10, hidden_layer_2, hidden_layer_3], -limit_3, limit_3))

#biases initializatrion
b0 = tf.Variable(tf.random_uniform([10, hidden_layer_1], -limit_1, limit_1))
b1 = tf.Variable(tf.random_uniform([10, hidden_layer_2], -limit_2, limit_2))
b2 = tf.Variable(tf.random_uniform([10, hidden_layer_3], -limit_3, limit_3))

q_x = tf.placeholder("float32", shape=[None, 10, input_features])
d_x = tf.placeholder("float32", shape=[None, 10, input_features])
y = tf.placeholder("float32", shape=[None, 10])

tensorflow

share|improve this question

asked Mar 25 at 6:09

Raju KomatiRaju Komati

34

• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Is something preventing you from flattening the inputs? If yes, then you will need to read math.stackexchange.com/questions/63074/… because without these concepts you will struggle to maintain the shape of the matrix
  
  – anand_v.singh
  Mar 25 at 6:19
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  no it's not about flattening
  
  – Raju Komati
  Mar 25 at 6:26
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  If you are determined to use a 3D tensor, then you need to learn about tensor contraction and manage your layer dimensions accordingly.
  
  – anand_v.singh
  Mar 25 at 6:36
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Weight matrices and bias vectors do not have anything do with your sample size. It depends on model parameters.
  
  – ARAT
  Mar 25 at 15:10
  

  
  
  
  

add a comment | 

0

training set x_train contain 10k examples and each input is of shape(10,23201)
and when i try to send the total training set for training it giving an error
but when i send 10 examples at a time it working fine.
How to change the code that will take all examples at once(what shape of weights will change my probllem)?

input_features = 23201
hidden_layer_1 = 300
hidden_layer_2 = 300
hidden_layer_3 = 128

limit_1 = tf.cast(np.sqrt(6/(input_features+hidden_layer_1)), tf.float32)
limit_2 = tf.cast(np.sqrt(6/(hidden_layer_1+hidden_layer_2)), tf.float32)
limit_3 = tf.cast(np.sqrt(6/(hidden_layer_2+hidden_layer_3)), tf.float32)
#weights initialization
w0 = tf.Variable(tf.random_uniform([10, input_features, hidden_layer_1], -limit_1, limit_1))
w1 = tf.Variable(tf.random_uniform([10, hidden_layer_1, hidden_layer_2],-limit_2, limit_2))
w2 = tf.Variable(tf.random_uniform([10, hidden_layer_2, hidden_layer_3], -limit_3, limit_3))

#biases initializatrion
b0 = tf.Variable(tf.random_uniform([10, hidden_layer_1], -limit_1, limit_1))
b1 = tf.Variable(tf.random_uniform([10, hidden_layer_2], -limit_2, limit_2))
b2 = tf.Variable(tf.random_uniform([10, hidden_layer_3], -limit_3, limit_3))

q_x = tf.placeholder("float32", shape=[None, 10, input_features])
d_x = tf.placeholder("float32", shape=[None, 10, input_features])
y = tf.placeholder("float32", shape=[None, 10])

tensorflow

share|improve this question

asked Mar 25 at 6:09

Raju KomatiRaju Komati

34

• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Is something preventing you from flattening the inputs? If yes, then you will need to read math.stackexchange.com/questions/63074/… because without these concepts you will struggle to maintain the shape of the matrix
  
  – anand_v.singh
  Mar 25 at 6:19
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  no it's not about flattening
  
  – Raju Komati
  Mar 25 at 6:26
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  If you are determined to use a 3D tensor, then you need to learn about tensor contraction and manage your layer dimensions accordingly.
  
  – anand_v.singh
  Mar 25 at 6:36
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Weight matrices and bias vectors do not have anything do with your sample size. It depends on model parameters.
  
  – ARAT
  Mar 25 at 15:10
  

  
  
  
  

add a comment | 

training set x_train contain 10k examples and each input is of shape(10,23201)
and when i try to send the total training set for training it giving an error
but when i send 10 examples at a time it working fine.
How to change the code that will take all examples at once(what shape of weights will change my probllem)?

input_features = 23201
hidden_layer_1 = 300
hidden_layer_2 = 300
hidden_layer_3 = 128

limit_1 = tf.cast(np.sqrt(6/(input_features+hidden_layer_1)), tf.float32)
limit_2 = tf.cast(np.sqrt(6/(hidden_layer_1+hidden_layer_2)), tf.float32)
limit_3 = tf.cast(np.sqrt(6/(hidden_layer_2+hidden_layer_3)), tf.float32)
#weights initialization
w0 = tf.Variable(tf.random_uniform([10, input_features, hidden_layer_1], -limit_1, limit_1))
w1 = tf.Variable(tf.random_uniform([10, hidden_layer_1, hidden_layer_2],-limit_2, limit_2))
w2 = tf.Variable(tf.random_uniform([10, hidden_layer_2, hidden_layer_3], -limit_3, limit_3))

#biases initializatrion
b0 = tf.Variable(tf.random_uniform([10, hidden_layer_1], -limit_1, limit_1))
b1 = tf.Variable(tf.random_uniform([10, hidden_layer_2], -limit_2, limit_2))
b2 = tf.Variable(tf.random_uniform([10, hidden_layer_3], -limit_3, limit_3))

q_x = tf.placeholder("float32", shape=[None, 10, input_features])
d_x = tf.placeholder("float32", shape=[None, 10, input_features])
y = tf.placeholder("float32", shape=[None, 10])

tensorflow

share|improve this question

asked Mar 25 at 6:09

Raju KomatiRaju Komati

34

input_features = 23201
hidden_layer_1 = 300
hidden_layer_2 = 300
hidden_layer_3 = 128

limit_1 = tf.cast(np.sqrt(6/(input_features+hidden_layer_1)), tf.float32)
limit_2 = tf.cast(np.sqrt(6/(hidden_layer_1+hidden_layer_2)), tf.float32)
limit_3 = tf.cast(np.sqrt(6/(hidden_layer_2+hidden_layer_3)), tf.float32)
#weights initialization
w0 = tf.Variable(tf.random_uniform([10, input_features, hidden_layer_1], -limit_1, limit_1))
w1 = tf.Variable(tf.random_uniform([10, hidden_layer_1, hidden_layer_2],-limit_2, limit_2))
w2 = tf.Variable(tf.random_uniform([10, hidden_layer_2, hidden_layer_3], -limit_3, limit_3))

#biases initializatrion
b0 = tf.Variable(tf.random_uniform([10, hidden_layer_1], -limit_1, limit_1))
b1 = tf.Variable(tf.random_uniform([10, hidden_layer_2], -limit_2, limit_2))
b2 = tf.Variable(tf.random_uniform([10, hidden_layer_3], -limit_3, limit_3))

q_x = tf.placeholder("float32", shape=[None, 10, input_features])
d_x = tf.placeholder("float32", shape=[None, 10, input_features])
y = tf.placeholder("float32", shape=[None, 10])

share|improve this question

asked Mar 25 at 6:09

Raju KomatiRaju Komati

34

share|improve this question

asked Mar 25 at 6:09

Raju KomatiRaju Komati

34

asked Mar 25 at 6:09

Raju KomatiRaju Komati

34

asked Mar 25 at 6:09

Raju KomatiRaju Komati

34