Closed mlipkovich closed 7 years ago
Hi Mikhail,
Thanks for reading the code and making a contribution. Although, I'm having a difficulty trying to understand your suggestion. tf.get_variable("variable_name", shape=[dim1, dim2]) is a function for creating/reusing a trainable variable in TensorFlow. The "variablename" can be anything you want, as long as you use it consistently. So could you explain why you are suggesting we change from 'W'+str(i) to 'W_'+str(i+1)? Also it's not so clear to me why changing that would give better performance.
Thanks, Ed
Hi Ed, Thank you for the response. After your comment I realized that my changes doesn't make sense. I missed that variables in the loop above are called "W_i" while variable that I was trying to change was "Wi" (with no underscore). So I thought that you were creating a new variable with the same name. Regarding to performance I meant quality - probably it happened due to some random perturbations. It is quite unstable for my dataset. Sorry for the disturbance. I'm closing the PR
Thanks, Mikhail
Hi, If I'm not mistaken that index should be updated in the following way. Performance seems to be better with this change