Getting Assertion Error when training data in neural network in python?

Question

I have a file BCICIV1bAF3.dat which contain data. The file size is 20x1

This is my code...

In newff function the range i decide based on Min/Max but i dont know how to decide the other parameters. How much hidden layer do i want etc.

import numpy as np
import neurolab as nl

input_data = np.fromfile('BCICIV1bAF3.dat' ,dtype=float)

print(len(input_data))
transformed_input_data = [[x] for x in input_data] # added
print(len(transformed_input_data))
output_data = np.fromfile('test.dat',dtype=float)

transformed_output_data = [[x] for x in output_data] # added

net = nl.net.newff([[-21, -10.5]], [1020, 1])
error = net.train(transformed_input_data, transformed_output_data)
predicted_output = net.sim(input_data)

Input Data:

-10.5 -91.7 -219.8 -227 -190.8 -218.7 -208.2 -205 -214.3 -202 -211.5 -211.1 -208.2 -212.4 -206 -206.7 -211.5 -210.7 -212 -215.1

Output Data:

-5.2 -45.6 -108.6 -112 -94.5 -106.7 -99.6 -98.5 -105.4 -101.2 -106.4 -106.5 -102.4 -105.7 -104 -97.9 -99.5 -101.3 -100.6 -103.7

Error:

Traceback (most recent call last):
  File "NNwork2.py", line 15, in <module>

     error = net.train(transformed_input_data, transformed_output_data)
  File "C:\Python34\lib\site-packages\neurolab\core.py", line 328, in __call__
    assert target.shape[0] == input.shape[0]
AssertionError

How can i train? And simulate the input_data?

If anyone could guide...I'll be very grateful. Thanks

From the content of your error message I can derive that the shape of your input is not what you expect. Maybe take a closer look on that. — PinkFluffyUnicorn
– PinkFluffyUnicorn, Commented Mar 27, 2017 at 11:52

pocza · Accepted Answer · 2017-03-28 17:56:44Z

2

Did you try other training methods? I saw in other answer that it helped, because of a bug in library. Available methods: train_gd, train_gdm, train_gda, train_gdx, train_rprop, train_bfgs (DEFAULT), train_cg

~~You can change it by calling:~~

~~net.trainf = nl.train.train_gd~~

If you could provide input data (even with changed values) it would be great.

I tried calling train method for input in form: [0,1,2,3...18,19] and it failed - I had to change input (and target) to [[0],[1],...[18],[19]]

EDIT:

Your data is in wrong format, you should transform it to list of lists. I don't have scipy on my machine, but try this:

import numpy as np
import neurolab as nl

input_data = np.fromfile('BCICIV1bAF3.dat' ,dtype=float)
transformed_input_data = [[x] for x in input_data] # added

print(len(transformed_input_data)) # changed
net = nl.net.newff([[-215.1, -10.5]], [20, 1])
error = net.train(transformed_input_data, transformed_input_data, epochs=500) # changed

EDIT 2:

I won't explain what neural network is (I didn't use them in quite a while), but it should look like this when we want to transform 3D input into 2D output with use of 1 hidden layer:

INPUT [3D] | HIDDEN LAYER | OUTPUT [2D]
                ----
               | H1 |
                ----
   ----
  | X1 |
   ----
                ----         ----
               | H2 |       | Y1 |
                ----         ----


   ----               
  | X2 |                      
   ----                      

                ----         ----
               | H3 |       | Y2 |
                ----         ----
   ----
  | X3 |
   ----
                ----
               | H4 |
                ----

Every X is multiplied by every H and we calculate output. How do we have those H values? They're computed by algorithms during training of neural network. We specify how many hidden layers we want and by trial and error arrive at satisfactory solution. Very important - we should use different data to train and to check ouput of neural network.

When could we use this particular network? E.g. when calculating how many Big Macs and fries people order at McDonald based on age, salary of client and placement of particular restaurant. It would look like this:

    -----
   | AGE |
    -----
                ----         ----------
               | H2 |       | BIG MACS |
                ----         ----------


   --------               
  | SALARY |                      
   --------                      

                ----         -----------
               | H3 |       |   FRIES   |
                ----         -----------
    -------
   | PLACE |
    -------
                ----
               | H4 |
                ----

So we could say that transformation looks like this f([Age, Salary, Place]) = [Big Macs, Fries]. We may have millions of input and output data records gathered by employees to train our network, so translating into python it would be list of inputs (3D) and we expect list of outputs (2D). E.g. f([[A_1, S_1, P_1], [A_2, S_2, P_2], ... , [A_N, S_N, P_N]]) -> [[BM_1, F_1], [BM_2, F_2], ... , [BM_N, F_N]]

We want same thing with your data, BUT we want to have both input and output to be 1D, hence we had to "wrap" every element of a list into another list. Same thing with output AND simulation input - you forgot that.

predicted_output = net.sim(input_data) # this won't work! You should wrap it

But testing neural network on training data is just wrong - you shouldn't do that

edited Mar 28, 2017 at 17:56

answered Mar 27, 2017 at 11:57

pocza

7206 silver badges10 bronze badges

Sign up to request clarification or add additional context in comments.

8 Comments

pocza Over a year ago

I'm glad it helped!

pocza Over a year ago

@Uzair - did you unchecked my answer?

Uzair Over a year ago

i checked it worked but i modified a little bit can you please take a look at that too. I am getting same error Thanks

pocza Over a year ago

@Uzair It is the same error, because you did the same mistake, but in different function. Is is not the issue with train, but with sim. You passed input in wrong dimensions - simulation function expects list ([] ) of vector inputs and in your case it should be 1D vector. Same thing was with training your network - functions wanted to have list of vectors. I can explain it better in my answer.

Uzair Over a year ago

I tried but getting this.... Traceback (most recent call last): File "NNwork2.py", line 42, in <module> error = net.train(transformed_input_data, transformed_output_data) File "C:\Python34\lib\site-packages\neurolab\core.py", line 165, in train return self.trainf(self, *args, **kwargs) File "C:\Python34\lib\site-packages\neurolab\core.py", line 328, in call assert target.shape[0] == input.shape[0] AssertionError

|

Collectives™ on Stack Overflow

Getting Assertion Error when training data in neural network in python?

1 Answer 1

8 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

8 Comments

Your Answer

Sign up or log in

Post as a guest

Related