Ggtcf

How do you say “my friend is throwing a party, do you wanna come?” in german

A peculiar integral identity

Was it really inappropriate to write a pull request for the company I interviewed with?

Can the Shape Water Cantrip be used to manipulate blood?

Are small insurances worth it

Has Wakanda ever accepted refugees?

Lock enemy's y-axis when using Vector3.MoveTowards to follow the player

Plagiarism of code by other PhD student

Misplaced tyre lever - alternatives?

Correct physics behind the colors on CD (compact disc)?

Ahoy, Ye Traveler!

I encountered my boss during an on-site interview at another company. Should I bring it up when seeing him next time?

How to merge row in the first column in LaTeX

Is there a math equivalent to the conditional ternary operator?

3.5% Interest Student Loan or use all of my savings on Tuition?

Why are special aircraft used for the carriers in the United States Navy?

Create chunks from an array

Why do phishing e-mails use faked e-mail addresses instead of the real one?

is 'sed' thread safe

Called into a meeting and told we are being made redundant (laid off) and "not to share outside". Can I tell my partner?

Make me a metasequence

Did Amazon pay $0 in taxes last year?

When was drinking water recognized as crucial in marathon running?

How do we objectively assess if a dialogue sounds unnatural or cringy?

keras constraint doesn't get called

1

Keras constraint docs: https://keras.io/constraints/.
I used NonNeg constraint to keep some of my network dense layer weights positive, usage like :

layer1 = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=NonNeg())

But some of the layer weights is still negative, I viewed the source code of keras constraint: https://github.com/keras-team/keras/blob/master/keras/constraints.py#L62 , it's very straightforward. So I wonder if the constraint is not called. So I wrote a customed constraint which throw an exception directly as

class ConstraintDemo(Constraint):
 def __init__(self):
 logging.info("CONSTRAINT called")
 def __call__(self, w):
 raise RuntimeError("GLQ ERROR")
 logging.info("CONSTRAINT called")
 return tf.cast(tf.greater_equal(w, 0.), tf.float32)

And I called the class similarly by

layer = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=ConstraintDemo())

Only the init method get called.The call method should raise a RuntimeError and stop the training process. But it didn't.

Why?

tensorflow version: 1.12.0

keras version:2.0.6

tensorflow keras

share|improve this question

asked 16 hours ago

richard.grichard.g

2,3251920

• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Please include the full model that includes this layer
  
  – Matias Valdenegro
  15 hours ago
  

  
  
  
  

add a comment | 

1

Keras constraint docs: https://keras.io/constraints/.
I used NonNeg constraint to keep some of my network dense layer weights positive, usage like :

layer1 = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=NonNeg())

But some of the layer weights is still negative, I viewed the source code of keras constraint: https://github.com/keras-team/keras/blob/master/keras/constraints.py#L62 , it's very straightforward. So I wonder if the constraint is not called. So I wrote a customed constraint which throw an exception directly as

class ConstraintDemo(Constraint):
 def __init__(self):
 logging.info("CONSTRAINT called")
 def __call__(self, w):
 raise RuntimeError("GLQ ERROR")
 logging.info("CONSTRAINT called")
 return tf.cast(tf.greater_equal(w, 0.), tf.float32)

And I called the class similarly by

layer = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=ConstraintDemo())

Only the init method get called.The call method should raise a RuntimeError and stop the training process. But it didn't.

Why?

tensorflow version: 1.12.0

keras version:2.0.6

tensorflow keras

share|improve this question

asked 16 hours ago

richard.grichard.g

2,3251920

• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Please include the full model that includes this layer
  
  – Matias Valdenegro
  15 hours ago
  

  
  
  
  

add a comment | 

Keras constraint docs: https://keras.io/constraints/.
I used NonNeg constraint to keep some of my network dense layer weights positive, usage like :

layer1 = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=NonNeg())

But some of the layer weights is still negative, I viewed the source code of keras constraint: https://github.com/keras-team/keras/blob/master/keras/constraints.py#L62 , it's very straightforward. So I wonder if the constraint is not called. So I wrote a customed constraint which throw an exception directly as

class ConstraintDemo(Constraint):
 def __init__(self):
 logging.info("CONSTRAINT called")
 def __call__(self, w):
 raise RuntimeError("GLQ ERROR")
 logging.info("CONSTRAINT called")
 return tf.cast(tf.greater_equal(w, 0.), tf.float32)

And I called the class similarly by

layer = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=ConstraintDemo())

Only the init method get called.The call method should raise a RuntimeError and stop the training process. But it didn't.

Why?

tensorflow version: 1.12.0

keras version:2.0.6

tensorflow keras

share|improve this question

asked 16 hours ago

richard.grichard.g

2,3251920

layer1 = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=NonNeg())

class ConstraintDemo(Constraint):
 def __init__(self):
 logging.info("CONSTRAINT called")
 def __call__(self, w):
 raise RuntimeError("GLQ ERROR")
 logging.info("CONSTRAINT called")
 return tf.cast(tf.greater_equal(w, 0.), tf.float32)

layer = Dense(1, activation='sigmoid', use_bias=True, kernel_constraint=ConstraintDemo())

share|improve this question

asked 16 hours ago

richard.grichard.g

2,3251920

share|improve this question

asked 16 hours ago

richard.grichard.g

2,3251920

asked 16 hours ago

richard.grichard.g

2,3251920

asked 16 hours ago

richard.grichard.g

2,3251920