twmeggs / anfis

Python implementation of an Adaptive neuro fuzzy inference system

ANFIS

anfis is a Python implementation of an Adaptive Neuro Fuzzy Inference System.

This ANFIS package is essentially a Python refactoring of the R code created by the team a the BioScience Data Mining Group, the original documentaion of which can be found here:

http://www.bdmg.com.ar/?page_id=176

As an exmaple of an ANFIS system, this Python code works (install and run the tests.py script to see it fit the some test data) but there is much left to do in order to improve the project. This is very much an early beta version. My python code is cruddy and NOT at all idiomatic. Documentation and doc strings need large amounts of work.

All useful contributions to make this a better project will be happily received.

Contributions

If you would like to contribute, please see the Issues section for ideas about what most needs attention.

Features

• Currently the implementation will support the use of three types of membership function:

• gaussmf: Gaussian

• gbellmf: Generalized bell

• sigmf: Sigmoid

This naming is taken from scikit-fuzzy, a fuzzy logic toolkit for SciPy, which can be found here: https://github.com/scikit-fuzzy/scikit-fuzzy

Each input variable can have an arbitrary number and mix of these membership functions.

• A user can define the number of epochs that will be run

• The returned ANFIS object can plot training errors, fitted results and the current shape of its membership functions (pre or post training)

Installation

anfis may then be installed by running:

$ pip install anfis

Dependencies

• Python
• numpy
• scikit-fuzzy
• matplotlib

Quickstart

Install anfis and navigate to the location of anfis/tests.py

From the command line run:

python tests.py

Alternatively, from the same location launch ipython and run:

run tests.py

This will set up and fit an ANFIS model based on the data contained in 'trainingSet.txt', using 10 epochs. Plots of the fitting errors and the model predicted output are graphed.

Contact

For other questions, please contact [email protected].

anfis's Issues

how to plot "plotMF"?

I cannot plot the anfis.plotMF. I don't know what to use as x and as InputVar. I tried several things, but the error “only integer scalar arrays can be converted to a scalar index” occurs everytime. Has someone an example?

As you used 4 membership function for each variable in example model. Can we use 3 membership function for each variable in our model with same ANFIS library ?

As you used 4 membership function for each variable in example model. Can we use 3 membership function for each variable in our model with same ANFIS library ?

In my model input variables are 12 and output variable is 1, if I use 4 mf for each variable, then total rules becoming
rules = 4 ^12 = 16777216
Consequent Parameters = 16777216 * ( 12 + 1) = 218103808

So my query is Can we use 3 membership function for each variable in our model with same ANFIS library ?

One more query How many observation do we need at least to model this data ( Consequent Parameters = 218103808)

http://www.bdmg.com.ar/?page_id=176 link is not opening

As you told there is a document related to ANFIS at following link. This link is not working.
http://www.bdmg.com.ar/?page_id=176

Implementing ANFIS for multiple input and multiple output

Hello can you please help me in using ANFIS for multiple output. i have 7 inputs and 5 outputs. I have to predict these 5 output in one model using ANFIS. Any help in this is regard is highly appreciated.

AttributeError: module 'anfis' has no attribute 'ANFIS'

how to fix it?

---

AttributeError Traceback (most recent call last)
Cell In[7], line 6
1 mf = [[['gaussmf',{'mean':0.,'sigma':1.}],['gaussmf',{'mean':-1.,'sigma':2.}],['gaussmf',{'mean':-4.,'sigma':10.}],['gaussmf',{'mean':-7.,'sigma':7.}]],
2 [['gaussmf',{'mean':1.,'sigma':2.}],['gaussmf',{'mean':2.,'sigma':3.}],['gaussmf',{'mean':-2.,'sigma':10.}],['gaussmf',{'mean':-10.5,'sigma':5.}]]]
5 mfc = anfis.membershipfunction.MemFuncs(mf)
----> 6 anf = anfis.ANFIS(X, Y, mfc)
7 anf.trainHybridJangOffLine(epochs=20)
9 print (round(anf.consequents[-1][0],6))

AttributeError: module 'anfis' has no attribute 'ANFIS'

Update PyPI release

Project's readme says to install it with pip, yet version available on pypi is not up to date with upstream and crashes on literally every python3 installation. If possible, it could be nice to get newer release there

how to plot membership functions

I cannot plot the anfis.plotMF. I don't know what to use as x and as InputVar. I tried several things, but the error “only integer scalar arrays can be converted to a scalar index” occurs everytime. Has someone an example?

I have been try to use this function like this:

print("Plotting membership_function")
anf.plotMF(X,0)

But the results are so bad

whatever I set this InputVar as '0' or '1', the results just as bad as I show

No module named 'membership'

AppData\Local\Programs\Python\Python310\lib\site-packages\anfis_init_.py", line 2, in
from membership import membershipfunction
ModuleNotFoundError: No module named 'membership'

Gaussian mean and sigma

Hhow can i integrate the decision of the mean and sigma for Gaussian based on neuronal networks (I want to make it automatically). Because the main feature, as I andertand the Anfis, is that it uses neuronal approch to decide, how membership functions look like.

(How did you estimate your menas and sigmas?)
Thx for your help in advance.
Rgrds,
Leyla

Syntax error

Someone please fix this issue. I had some trouble using the membershipfunction, but then I did the following but still get an error:
On my notebook, I use import anfis and from anfis.membership import membershipfunction. The rest is as the current code on github is.
I get this error:
File "/opt/anaconda3/lib/python3.8/site-packages/anfis/membership/membershipfunction.py", line 19
if len(rowInput) <> len(self.MFList):
^
SyntaxError: invalid syntax

Different number of MF for each input variable

i define the MF for four of my input variables as below:
mf = [[['gaussmf',{'mean':13.45,'sigma':3.2}], ['gaussmf',{'mean':25.7,'sigma':3.2}]], [['gaussmf',{'mean':41.74,'sigma':6.3}], ['gaussmf',{'mean':54.53,'sigma':6.3}], ['gaussmf',{'mean':66.51,'sigma':6.3}]], [['sigmf',{'b':1014.26,'c':-0.6}], ['sigmf',{'b':1012.1,'c':0.6}]], [['sigmf',{'b':60,'c':-0.6}], ['gaussmf',{'mean':73.31,'sigma':7.3}], ['sigmf',{'b':88,'c':0.6}]]]

but resulted in error when i run anf.trainHybridJangOffLine(epochs=3) like below:

i don't know how to solve this. please help me @twmeggs

How do you determine sigma or mean value when using another dataset?

in the code you would find this

[
            ['gaussmf', {'mean': 0., 'sigma': 1.}],
            ['gaussmf', {'mean': -1., 'sigma': 2.}],
            ['gaussmf', {'mean': -4., 'sigma': 10.}],
            ['gaussmf', {'mean': -7., 'sigma': 7.}]
],

how do we calculate it for different dataset?

IndexError: list index out of range

Hello

I have a dataset with 3 features and one output but I face this problem and I do not know how to solve it?

Any idea?

Error:

Number of variables does not match number of rule sets

IndexError Traceback (most recent call last)
in ()
51 #################################################################
52
---> 53 anf.trainHybridJangOffLine(epochs=epochs_anfis)
54
55 ####################################################################

3 frames
in (.0)
22 print("Number of variables does not match number of rule sets")
23
---> 24 return [[self.funcDictself.MFList[i][k][0] for k in range(len(self.MFList[i]))] for i in range(len(rowInput))]

IndexError: list index out of range

Example won't work

Hi!

Thanks for your work, but I'm afraid your example isn't working. The error just keeps going up and up, even raising the epochs number. There is no going down...

Is there any way to fix this?

Thanks in advance!

No module named 'membership'

I installed the anfis package using anaconda prompt and pip. I am using python 3.6 and the error that appears is: "ModuleNotFoundError: No module named 'membership'".

What could be happening?

Thanks in advance!

PD: the membership folder is inside the anfis folder and contains all the functions. Furthermore, the anfis package is inside the anaconda lib folder.

clustering type

What kind of clustering algorithm does this system work with?
gridPartion, subClustering, or FCM? And where can these values be set?

Add scikit-fuzzy as official dependency, push new version to PyPi

Scikit-Fuzzy is now on PyPi, so it can be formally added as a dependency! setup.py and the README should be updated.

I'd be happy to do this, if you'd like me to make a PR.

Current error increasing with the number of epochs (from epoch 20)

Hi,
I would like to know if anyone has encountered this problem:
When using the function for a more complex problem, I noticed that the current model error starts to increase.
I thought it was an error in my program, but when I tested the repository example, when I increased the epochs a little more, I found the same problem.

I leave below the image referring to the results of the example of the repository for more than 20 epochs:

On the one hand, I don't know if I should let the model evolve a little in this sense, hoping that it is just trying not to overfit.
However, I think that the assumption is always that the error decreases or at least stabilizes.
I tried to "fix" this by putting a simple "if" inside the anfis.py function itself that when the error increases more than 0.001, for more than 3 times (for exemple), it stops the training loop. However, I don't know to what extent this cannot be detrimental to the results

Thanks in advance

python 3 compatability

Hi, just an fyi that the membership folder is coming out as nested in the anfis/anfis folder and so in order to get to run the membership folder needs to placed in the parent dir and imported separately to both the test and anfis.py files. It wont run otherwise.

How use multiple input in anfis ??

this code only 2 variable/parameter column and 1 target column, can i use for multiple variable/parameter ??

Documentation link is broken

Can it works on Python3

same as the question

Shouldn't we use fuzzy multiplication?

Hi. Normal multiplication is used with the np.product(x) fuction in the ANFIS codes .
Shouldn't it be necessary to use fuzzy multiplication and addition operations because membership degrees are used?
If it should be as used in the codes, can you write the reason?
This answer is so important for me. I would be very grateful if you could answer.

The system is available for thousands of time series?

I want to use the system to predict the CPU of VMs. But, I encounter a challenge that The number of VM is too many.

please drop imp!

in new update of python imp dropped and after run tests.py raise error ModuleNotFoundError: No module named 'imp'

please fix this,

thansk alot

Can this implementation be used for classification?

I am trying to use ANFIS to do multi-classification. Can this implementation work? Thanks in advance!