How to visualize ludwig library models learning curves programatically?

Question

I am using Uber's ludwig library to train neural network models. I want to programatically (via python code, using the documentation's example) visualize the learning curves of my models to analyze their performance using the following code:

import ludwig
ludwig.visualize.learning_curves(
  [train_stats],
  TARGET,
  model_names=None,
  output_directory=None,
  file_format='pdf'
)

However, I am obtaining the following error when running the code:

AttributeError: module 'ludwig' has no attribute 'visualize'

I know that I can generate the visualization via command line. However, I need to do it programatically.

score 0 · Answer 1 · answered Oct 19 '20 at 10:26

0

Do it like that.

from ludwig import visualize
visualize.learning_curves(
  [train_stats],
  TARGET,
  model_names=None,
  output_directory=None,
  file_format='pdf'
)

answered Oct 19 '20 at 10:26

shahid hamdam

751
1
10
24

score 0 · Accepted Answer · answered Dec 03 '20 at 17:20

I was able to run the same code without changing it. I just updated the version of ludwig but I am not sure if that was the solution to my problem. I kept using this piece of code and it worked:

import ludwig
ludwig.visualize.learning_curves(
  [train_stats],
  TARGET,
  model_names=None,
  output_directory=None,
  file_format='pdf'
)

How to visualize ludwig library models learning curves programatically?

2 Answers2