just starting
Neural Nets are trained by a set of training data.
As a minimum this training data has many sets of input and output.
The input and output are numbers, that may represent anything,
e.g. characters, animals, any objects, values, etc.

There must be the same number of inputs and same number of outputs
in every entry in the training data.
In a neural net, there is an input neuron for each input.
In a neural net, there is an output neuron for each output.
There are typically hidden layers of neurons between the
input layer and output layer.
The process of training the neural net is to determine the
weights that are multiplied by a neuron in one layer and
added into a neuron in the next layer.

A simple neural net shows four input neurons, three output neurons,
and two hidden layers. It is allowed, but not required, that each
layer only feeds the next layer, and that all neurons in the
previous layer contribute to all neurons in the next layer.


click to see one possible neural net

Above the inputs go into neurons N00, N01, N02, N03.
The outputs are available in neurons N30, N31, N32.

Example: N10 gets sum of
         N00*w000 + N01*w010 + N02*w020 + N03*w030

Example: N32 gets sum of
         N20*w202 + N21*w212 + N22*w222 + N23*w323

In software neurons have double subscript, layer and position.
In software weights have tripple subscript
                         from_layer, from_position, to_position
the next layer is implicit.
 
sample input data sets may have:

The number of layers in the neural net.
The number of neurons (positions) in each layer.
A fully connected neural net as shown above is used.
Possible optional random or set values

The number of data set entries,
the number of inputs in an entry,
the number of outputs in an entry.

There may be a comment, anything following a #

d_and.train


sample weights and test results coming soon
and

last updated  10/1/2016