Sep 4, 2018 · 1 min read
Nice article! I found the hint to joblib particularly helpful. I had been caching intermediate results manually and was wondering whether my practices would evolve into reusable patterns. The joblib caching looks like a reasonable solution.
Your example function in 10 should probably end with
return (X - mi) / (mx - mi + epsilon)otherwise you would scale values between +2 and +6 by dividing X-2 by 8 instead of 4.