-
Notifications
You must be signed in to change notification settings - Fork 137
Enhanced the random seed generation function. #328
Conversation
All the requested changes (except for the |
This should be ready for another look now. |
hashCopy[4].rotate(rightBy: 41) | ||
let ch = (hashCopy[4] & hashCopy[5]) ^ ((~hashCopy[4]) & hashCopy[6]) | ||
let t1 = hashCopy[7] &+ s1 &+ ch &+ k[j] &+ UInt64(M[j]) | ||
hashCopy[7] = hashCopy[6] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@stephentyrone Could you please advise?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I found this which seems relevant. :)
@eaplatanios, If you need any help with all these tasks you are undertaking, I'd love to help you out. I really like the enhancements we've had in the past week. So if there are trivial/non-trivial tasks you need some help with, do tag me. |
@Shashi456 Thanks a lot! Sorry for the late response but I was traveling. I don't have something planned right now, but I will definitely tag you if anything comes up. :) |
This PR enhances the function that generates default random seeds for TensorFlow. This is needed because the previous approach is weak in multi-process settings, which are quite common when training or evaluating machine learning models. The implementation is inspired from OpenAI Gym.
I also added an implementation on SHA512 in the utilities to avoid having any external library dependencies along with a test for the hash function. The SHA512 implementation is based on that of CryptoSwift.