How do you make a prediction (forecast) from a trained network for a given input?

Question

Below is my code for a neural network,with 3 inputs and 1 hidden layer and 1 output:

#Data 
ds = SupervisedDataSet(3,1)

myfile = open('my_file.csv','r')

for data in tf.myfile ():
   indata =  tuple(data[:3])
   outdata = tuple(data[3])
   ds.addSample(indata,outdata)

net = FeedForwardNetwork() 
inp = LinearLayer(3) 
h1 = SigmoidLayer(1) 
outp = LinearLayer(1)

# add modules 
net.addOutputModule(outp) 
net.addInputModule(inp) 
net.addModule(h1)

# create connections 
net.addConnection(FullConnection(inp, h1))  
net.addConnection(FullConnection(h1, outp))

# finish up 
net.sortModules()

# initialize the backprop trainer and train 
trainer = BackpropTrainer(net, ds)
trainer.trainOnDataset(ds,1000) trainer.testOnData(verbose=True)

print 'Final weights:',net.params

My question is,if you want to use this trained neural network to make a forecast based on specific inputs,how do you do it?

Answer 1

According to the documentation, you can test specific inputs with the activate method on your network. Assuming your input looks something like (1 2 3) your code would look like

net.activate((1,2,3))

Answer 2

If I understand you correctly, your data has a time order. What I do for making forecast is to shift the data table, in order to present the next output as target for the training. For example, if you have this kind of data:

w1 x1 y1 z1

w2 x2 y2 z2

w3 x3 y3 z3

w4 x4 y4 z4

. . .

and you want to predict z2, you construct a table like:

w1 x1 y1 z1 | z2

w2 x2 y2 z2 | z3

w3 x3 y3 z3 | z4

. . .

Then you present the last column as target for the training. Of course, you lose one line at the end of your table.

You can also improve the output by giving the difference between steps as an additional input (gives you the dynamical effect:

w2 x2 y2 z2 (w2-w1) (z2-z1) | z3

w3 x3 y3 z3 (w3-w2) (z3-z2) | z4

. . .