0

I am working on LSTM model and using cuda and cudnn in tensorflow. CUDA version is 11.2, cuDNN version is 8.1, tf version is 2.5 and python version is 3.9. I have to use costcla library of PyPi but it is compatible with python v3.5. How can I use costcla library without doing other changes? I am new to Deep Learning so need some suggestions

Does costcla give similar results as using sample_weight parameter in fit() of DL model?

Priyanka Goel
  • 1
  • 2
  • Hi @Priyanka Goel, There is no other way. You need to downgrade the TF version to 2.3 or below to be compatible with `Python 3.5` as well as also need to downgrade the other supported libraries (`CUDA 10.1, cuDNN 7.6`) to have GPU support with the Tensorflow. It seems both costcla and sample_weight gives similar results because costcla calculates the cost for a given class whereas sample_weight is a numpy array of weights for the training samples, used for weighting the loss function (during training only). – TF_Renu Patel Jul 27 '23 at 10:12
  • Thank you for the reply. I have one more query- will costsensitive package be helpful in case of highly imbalanced dataset? Sample_weight does not work on test data so its effect on scores is null – Priyanka Goel Jul 28 '23 at 19:08

0 Answers0