Talos: Hyperparameter Optimization for Keras and PyTorch.

Talos is a hyperparameter optimization package made for data scientists and data engineers who are tired of mindless parameter hopping and confusing optimization solutions that add complexity instead of reducing it. It works with any Keras, TensorFlow (tf.keras) or PyTorch model, takes minutes to implement, involves no new syntax to learn and adds zero new overhead to your workflow.

References:

Documentation
Important Notes

Interactive job
Interactive jobs should be used for debugging, graphics, or applications that cannot be run as batch jobs.

Allocate an interactive session and run the program. Sample session:

[user@biowulf]$  sinteractive --gres=gpu:k80:1,lscratch=5 
[user@cn4235 ~]$ module load talos   
[+] Loading singularity  3.8.5-1  on cn4235
[+] Loading cuDNN/7.6.5/CUDA-10.1 libraries...
[+] Loading CUDA Toolkit  10.1.105  ...
[+] Loading talos  1.0
Copy a test python script, iris.py, to your current foilder:
[user@cn4235 ~]$ cp $TALOS_TEST/* . 
Run the test script:
[user@cn4235 ~]$python-talos iris.py 
...
Successfully opened dynamic library libcublas.so.10
...
 11%|##6                     | 1/9 [00:07<01:02,  7.79s/it]
 22%|#####3                  | 2/9 [00:12<00:42,  6.10s/it]
 33%|########                | 3/9 [00:17<00:33,  5.55s/it]
 44%|##########6             | 4/9 [00:21<00:23,  4.79s/it]
 56%|#############3          | 5/9 [00:24<00:17,  4.37s/it]
 67%|################        | 6/9 [00:28<00:12,  4.13s/it]
 78%|##################6     | 7/9 [00:31<00:07,  3.86s/it]
 89%|#####################3  | 8/9 [00:35<00:03,  3.68s/it]
100%|########################| 9/9 [00:38<00:00,  4.27s/it]
End the interactive session:
[user@cn4235 ~]$ exit
salloc.exe: Relinquishing job allocation 46116226
[user@biowulf ~]$