‘Number of neurons sensitivity’ for non-binary input and output

21Jul10

The previous post showed an ANN with binary input and output . The input could be 0 or 1, the output could be 1 or -1.
In this post, we do similarly, but with a non-binary, but still discrete case.
The input is still the RUT %difference from its SMA(180).
The distribution of the next week average %up/down direction chart is still valid as it was posted in the last post.

The distribution of the next week average %gain chart is:

This two charts matches each other and can be explained by the spring model: if the RUT is too far from its MA, it tends to regress back.

We teach the NN.
The input is still 1 dimension. The RUT %difference from its SMA(180).
The output is the next week RUT %gain.

Let’s see the NN surface, how the NN predict this for different nNeurons.
The number of Neurons are: 1, 2, 3, 4, 5, 15.

We can laugh our head off looking at the prediction capability of the 15 neurons case.
The best predictor again is the 1 neuron or the 2 neuron case.
The 1 neuron case maybe doesn’t work here, because it never outputs negative values.
That is strange, but I accept it for now. (The ANN concept never promised to find the optimal weights. It can stuck in local minima.)
The conclusion is the same: the less the number of neurons, the better.

Advertisements


No Responses Yet to “‘Number of neurons sensitivity’ for non-binary input and output”

  1. Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: