CNN for angle recognition

[ChrisZeThird]

Summary

As explained in the main issue note #7, the FF cannot find the angle among the noise, even with the derivative of the diagrams, or when it's trained with synthetic data. Instead of a FF, we could use a CNN, much more accurate for this kind of problem. It seems to be a rational alternative considering all the options found while trying to solve the FF prediction problem.

[ChrisZeThird]

Here is a first test of training on experimental data

[ChrisZeThird]

I did not follow the order of the workflow but at least it's done. Experimental data for training also causes issue with CNN for the moment:

[ChrisZeThird]

Gaussian noise

I tried to add gaussian noise to synthetic data in various way (scipy, normal array distribution). Here are the results:

No noise, anti-alias line

No noise, normal distribution, anti-alias line

Gaussian noise, normal distribution, anti-alias line

Gaussian noise, normal distribution, no anti-alias

[ChrisZeThird]

Training on synthetic data

These results are temporary, as the current network is not the best probably.

[ChrisZeThird]

CNN will be on hold for now. A simple regression algorithm should be enough to solve the problem.