Use neuronetwork(tensorflow) for regression 0

This post is not a tutorial, but rather a logbook of what we attempted.

The learning logbook starts with using neutonetwork to do regression.

The data is manually generate using a very simple formula.

$$y=3*x_0+sin(x_1)$$, initially, we do not add any noise term.

accuracy function is below

In the first attempt, a single intermediate layer neuronetwork is applied.

It is observed that, with 60 hidden neurons and 25000 training steps, the prediction accuracy can be largely fluctuated according to the initialization value.

Fig 1. Prediction results with 60 hidden neurons, repeated 50 times.

Leave a Reply

Your email address will not be published. Required fields are marked *