I am pretty new in using python and neurolab and I have a problem with the training of my feed forward neural network. I have built the net as following: net = nl.net.newff([[-1,1]]*64, [60,1]) net.init() testerr = net.train(InputT, TargetT, epochs=100, show=1) and my target output is a vector between 0 and 4. When I use the nl.train.train_bfgs I have in the console: testerr = net.train(InputT, TargetT, epochs=10, show=1) Epoch: 1; Error: 55670.4462766; Epoch: 2; Error: 55649.5; As you can see, I fixed the number of epochs to 100 but it stops at the second epoch and after the test of the net with Netresults=net.sim(InputCross) I have as test output array a vector of 1 (totally wrong). If I use the other training functions I have the same output testing vector full of 1 but in that case during the training, the epochs reach the number that I set but the error displayed doesn't change between the epochs. The same if the target output vector is between -1 and 1. Any suggestion? Thank you very much!

What steps will reproduce the problem?
1. training nff
2. large input --42 input neurons floaing point vales as inputs
3. out put (0,0,0,1)  ,(0,0,1,0),(0,1,0,0),(1,0,0,0)

What is the expected output? What do you see instead?
   decrease in error but constant error seen

What version of the product are you using? On what operating system?
   most recent version on my fedora 17(linux 3.7)

Please provide any additional information below.

This was to recognize handgesture where the inputs are x,y,z acceleration 
values and 4 patterns are the possible outputs

Original issue reported on code.google.com by sarath.sp06 on 10 Mar 2013 at 5:56

Attachments:

• ann.py
• pro.tar.gz

What steps will reproduce the problem?
1. Import neurolab as nl and create a simple vector n with 3 values, one value 
< 0, one value > 0 and one value < 0, e.g. n = (-0.5763, 0.8345, -0.1234)
2. let f = nl.trans.Competitive() 
3. a = f(n)

What is the expected output? What do you see instead?
I expect [0, 1, 0], instead I see [1, 0, 0 ]

What version of the product are you using? On what operating system?
Version 0.1.0 on Ubuntu 11.04

Please provide any additional information below.

Original issue reported on code.google.com by [email protected] on 5 Aug 2011 at 7:21

What steps will reproduce the problem?
Using the neurolab Net modules is not possible to parallelize its execution due 
to the unbound functions are unpickable through queues.

It could be solved not defining unbound functions of trainf or errof during the 
instantiation of the Net structure. Its just a fast solution, defininf the 
errorf and trainf as strings and calling them in the right moment as in error 
function calling:

return getattr(error, net.errof)(target - output)

Original issue reported on code.google.com by [email protected] on 6 Mar 2014 at 1:13

I have added l2 and l1 regularization capabilities for training feed-forward 
networks to the Neurolab library. I have also added cross-entropy error (used 
in logistic regression) to error.py.

I would like to see these modifications incorporated into the standard 
distribution of Neurolab, but I'm not sure how to contact the developers of 
Neurolab. If anyone has advice, please let me know.

You can find my modifications, an explanation, and a demonstration here:
1. https://github.com/kwecht/NeuroLab
2. 
http://nbviewer.ipython.org/github/kwecht/NeuroLab/blob/master/Adding%20Regulari
zation%20to%20Neurolab.ipynb

Original issue reported on code.google.com by [email protected] on 22 Dec 2014 at 9:18

The package installs automatically and successfully into the python 2.7 
directories...I am using python 3.2 and cannot import it into a python 3.2 
program,  I get a "no module found" error.   When I set the path to the 2.7 
directory and try to 'import neurolab' it finds _init_.py and then pulls an 
error saying no 'net' module found.  I admit I am new to python and linux so 
any help that can be given is appreciated. I have installed this package using 
easy_install, pip and the setup.py methods and get the same result.

Original issue reported on code.google.com by [email protected] on 2 Jan 2014 at 3:29

Hello everyone!

I've been trying to add the Levenberg-Marquardt algorithm as a training-method 
but I fail doing so - I get a strange error-message I don't understand.

Here's the small method I've added to spo.py, using scipy.optimize.leastsq:
class TrainLM(TrainSO):
    def __call__(self, net, input, target):
        from scipy.optimize import leastsq
        x = leastsq(self.fcn, self.x.copy())
        self.x[:] = x

And in train/__init__.py I added the line train_lm = trainer(spo.TrainLM)

A simple test-run using the XOR-function looks like this:
>>import neurolab
>>target = [[0], [1], [1], [0]]
>>input = [[0,0], [0,1], [1,0], [1,1]]
>>net = neurolab.net.newff([[-0.5, 0.5], [-0.5, 0.5]], [5, 1])
>>net.trainf = neurolab.train.train_lm
>>err = net.train(input, target, show=15)

I get this error-message:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.7/dist-packages/neurolab/core.py", line 163, in train
    return self.trainf(self, *args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/neurolab/core.py", line 345, in __call__
    train(net, *args)
  File "/usr/local/lib/python2.7/dist-packages/neurolab/train/spo.py", line 44, in __call__
    x = leastsq(self.fcn, self.x.copy())
  File "/usr/lib/python2.7/dist-packages/scipy/optimize/minpack.py", line 278, in leastsq
    raise TypeError('Improper input: N=%s must not exceed M=%s' % (n,m))
TypeError: Improper input: N=21 must not exceed M=1

I've been playing around with this and I do not understand what M is in this 
case? What am I doing wrong? 

Thanks in advance!

Original issue reported on code.google.com by [email protected] on 6 Nov 2011 at 5:52

А можно увидеть пример работы сети Элмана 
для предсказания числовых 
последовательностей ?
Обучив сеть на последовательности [1, 3, 5, 7, 
9], и подав на вход, я хочу чтобы она 
предсказала следующий элемент (11) ? 
Пробежавшись по диагонали по исходникам не 
смог найти ответа - step() требует элемент и 
возвращает примерно его же.

Original issue reported on code.google.com by [email protected] on 17 Oct 2011 at 11:02

Hi. Your doc pages:

https://pythonhosted.org/neurolab/ex_newff.html
https://pythonhosted.org/neurolab/ex_newelm.html

have the same example.

The newelm example is missing.

Original issue reported on code.google.com by [email protected] on 20 Aug 2014 at 3:35

First, thanks for this great project.
I would like to cite neurolab in my papers. Is there any paper about neurolab 
itself?

Thank you.

Original issue reported on code.google.com by [email protected] on 1 May 2014 at 8:19

What steps will reproduce the problem?
1.I installed by:pip install neurolab
2.Run the Single Layer Perceptron (newp)


What is the expected output? What do you see instead?
The output is:
Traceback (most recent call last):
  File "neurolab.py", line 10, in <module>
    import neurolab
  File "/home/lilei/neurolab.py", line 17, in <module>
    net = neurolab.net.newp([[0, 1],[0, 1]], 1)
AttributeError: 'module' object has no attribute 'net'

And if I copy the neurolab file to the path of the example file,then the error 
is fixed.
What version of the product are you using? On what operating system?
0.2.3

Please provide any additional information below.

Original issue reported on code.google.com by [email protected] on 26 Jun 2013 at 4:32

While using function newlvq in net, I encountered the following error: line 184, in newlvq layer_out.np['w'][n][st:i].fill(1.0) TypeError: slice indices must be integers or None or have an index method

on further investigation, I found that this relates to the solution #481.

The array produced by line 181 inx = np.floor(cn0 * pc.cumsum()) needs to be changed from float64 to integers to be used as the i inputs in line 184.

while I would not call this an elegant solution, it does resolve the type issue:

def newlvq(minmax, cn0, pc):

pc = np.asfarray(pc)
assert sum(pc) == 1
ci = len(minmax)
cn1 = len(pc)
assert cn0 > cn1

layer_inp = layer.Competitive(ci, cn0)
layer_out = layer.Perceptron(cn0, cn1, trans.PureLin())

layer_out.initf = None
layer_out.np['b'].fill(0.0)
layer_out.np['w'].fill(0.0)
inx = np.floor(cn0 * pc.cumsum())

Modification begins

**v = 0
new = []
for i in inx:
    new.append(int(i))
    v += 1**
for n, i in enumerate(**new**):
    st = 0 if n == 0 else **new**[n - 1]
    layer_out.np['w'][n:st:i].fill(1.0)

End of modification

net = Net(minmax, cn1, [layer_inp, layer_out],
                        [[-1], [0], [1]], train.train_lvq, error.MSE())

return net

Thanks, Jonathan

From this page: https://pythonhosted.org/neurolab/ex_newlvq.html Using numpy 1.13.3 and neurolab 0.3.5 The error message I got was:

Traceback (most recent call last):
  File "learning_vector.py", line 20, in <module>
    net = nl.net.newlvq(nl.tool.minmax(input), 4, [.6, .4])
  File "/usr/local/lib/python2.7/dist-packages/neurolab/net.py", line 179, in newlvq
    layer_out.np['w'][n][st:i].fill(1.0)
TypeError: slice indices must be integers or None or have an __index__ method

I fixed the issue by editing the neurolab/net.py and changing the line 127 from layer_out.np['w'][n][st:i].fill(1.0) to layer_out.np['w'][n][int(st):int(i)].fill(1.0)

Thanks!

Martin Rioux

Why 'epochf' is called before 'learn' in train/gd.py algorithms?

Imagine you need only one epoch (for example, in reinforcement learning's step). When you specify 'epochs=1' the algorithms stop before any learning. When you specify 'epochs=2' you get additional unnecessary gradient calculation (so that you double the expensive calculations count).