Trjtdtk

Why is so much work done on numerical verification of the Riemann Hypothesis?

Biological Blimps: Propulsion

Electoral considerations aside, what are potential benefits, for the US, of policy changes proposed by the tweet recognizing Golan annexation?

How to explain what's wrong with this application of the chain rule?

What should you do when eye contact makes your subordinate uncomfortable?

Drawing ramified coverings with tikz

Character escape sequences for ">"

Fear of getting stuck on one programming language / technology that is not used in my country

The Staircase of Paint

Why did the HMS Bounty go back to a time when whales are already rare?

How could a planet have erratic days?

What prevents the use of a multi-segment ILS for non-straight approaches?

What is Cash Advance APR?

Multiplicative persistence

What is this called? Old film camera viewer?

Writing bit difficult equation in latex

Which one is correct as adjective “protruding” or “protruded”?

By means of an example, show that P(A) + P(B) = 1 does not mean that B is the complement of A.

Problem with TransformedDistribution

Why should universal income be universal?

What if a revenant (monster) gains fire resistance?

Why do we read the Megillah by night and by day?

Earnshaw’s Theorem and Ring of Charge

How to implement a feedback to keep the DC gain at zero for this conceptual passive filter?

Python Library for Neural Networks (no Tensors)

Best python library for neural networksBest Julia library for neural networksPython Neural Network Library - With Dynamic TopologiesPython Neural Network library/program with simple installation? (Solved?)Neural Networks overfittingNeural network only converges when data cloud is close to 0Regression with Neural Networks in Tensorflow problemRBF neural network python library/implementationMAE and MSE are Nan for regression with Neural Networks?Loss function minimizing by pushing precision and recall to 0

0

$begingroup$

I’m having an absolute nightmare with Keras and TF and I think by now it’s time to attempt a different approach with a different library (plan C is to build the network from scratch)

My neural network needs to have n regression outputs, let’s say 3, after a couple of densely connected layers. Input does not matter, I input a row of 0’s.

The reason it doesn’t matter is that my fitness function is custom as follows:

loss = 100 - get_accuracy(a, b, c)

Where a, b and c are the three numerical outputs given by the network at that time. Thus, the network is taking the inverse of accuracy as loss to in effect maximise it.

Is there a Python library that will let me implement this with ease? Keras was perfect up until I realised the python code within a custom loss is only executed once per compile and not once per step.

(I know there are better approaches to maximise this function, I want to do this to compare to those approaches)

machine-learning python neural-network regression

share|improve this question

asked Mar 19 at 12:04

Jordan BirdJordan Bird

1

New contributor

Jordan Bird is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.

$endgroup$

• 
  
  
  

  
  

  
  
  
  
  
  

  
  $begingroup$
  "I realised the python code within a custom loss is only executed once per compile and not once per step." -- this should not happen. Function is called after each batch (for updating weights).
  $endgroup$
  – Shamit Verma
  Mar 19 at 13:41
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  $begingroup$
  If I have a global variable of 100 and the loss function decrements it, it always ends as 99 no matter how many epochs it runs for.
  $endgroup$
  – Jordan Bird
  Mar 19 at 20:22
  

  
  
  
  

add a comment | 

0

$begingroup$

I’m having an absolute nightmare with Keras and TF and I think by now it’s time to attempt a different approach with a different library (plan C is to build the network from scratch)

My neural network needs to have n regression outputs, let’s say 3, after a couple of densely connected layers. Input does not matter, I input a row of 0’s.

The reason it doesn’t matter is that my fitness function is custom as follows:

loss = 100 - get_accuracy(a, b, c)

Where a, b and c are the three numerical outputs given by the network at that time. Thus, the network is taking the inverse of accuracy as loss to in effect maximise it.

Is there a Python library that will let me implement this with ease? Keras was perfect up until I realised the python code within a custom loss is only executed once per compile and not once per step.

(I know there are better approaches to maximise this function, I want to do this to compare to those approaches)

machine-learning python neural-network regression

share|improve this question

asked Mar 19 at 12:04

Jordan BirdJordan Bird

1

New contributor

Jordan Bird is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.

$endgroup$

• 
  
  
  

  
  

  
  
  
  
  
  

  
  $begingroup$
  "I realised the python code within a custom loss is only executed once per compile and not once per step." -- this should not happen. Function is called after each batch (for updating weights).
  $endgroup$
  – Shamit Verma
  Mar 19 at 13:41
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  $begingroup$
  If I have a global variable of 100 and the loss function decrements it, it always ends as 99 no matter how many epochs it runs for.
  $endgroup$
  – Jordan Bird
  Mar 19 at 20:22
  

  
  
  
  

add a comment | 

$begingroup$

I’m having an absolute nightmare with Keras and TF and I think by now it’s time to attempt a different approach with a different library (plan C is to build the network from scratch)

My neural network needs to have n regression outputs, let’s say 3, after a couple of densely connected layers. Input does not matter, I input a row of 0’s.

The reason it doesn’t matter is that my fitness function is custom as follows:

loss = 100 - get_accuracy(a, b, c)

Where a, b and c are the three numerical outputs given by the network at that time. Thus, the network is taking the inverse of accuracy as loss to in effect maximise it.

Is there a Python library that will let me implement this with ease? Keras was perfect up until I realised the python code within a custom loss is only executed once per compile and not once per step.

(I know there are better approaches to maximise this function, I want to do this to compare to those approaches)

machine-learning python neural-network regression

share|improve this question

asked Mar 19 at 12:04

Jordan BirdJordan Bird

1

New contributor

Jordan Bird is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.

$endgroup$

share|improve this question

asked Mar 19 at 12:04

Jordan BirdJordan Bird

1

New contributor

Jordan Bird is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.

share|improve this question

asked Mar 19 at 12:04

Jordan BirdJordan Bird

1

New contributor

Jordan Bird is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.

asked Mar 19 at 12:04

Jordan BirdJordan Bird

1

New contributor

Jordan Bird is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.

asked Mar 19 at 12:04

Jordan BirdJordan Bird

1