One-variable function approximation with four-layer DNN

Overview

In order to understand Chainer's Optimizer, I wrote a one-variable function approximation program with 4-layer DNN, referring to this. For some reason, this program did not perform forward calculation with the call function that inherited Chain, so I decided to fix it. I put the program I fixed on my github. I tried sin, exp, sqrt, abs, sqrt (abs (sin (exp (x))).

result

exp Good vibes. sin There is a place that is blurry. sqrt Good vibes. abs It feels good contrary to expectations. sqrt(abs(sin(exp(x))) I feel motivated.

What I thought

The development environment is terribly difficult to build (due to Python's packaging system). Not only the first time, but it is troublesome to start every time.