Hi, in the first example in the README, it states:

"For example, MDR can be used to construct a new feature composed from two existing features:"

but "GAMETES_Epistasis_2-Way_20atts_0.4H_EDM-1_1" used in the example has 21 columns, not 2.

The resulting output is a single column, which is a single feature -- is it that there's a single feature produced because that's what those 21 columns boiled down to, or is it because only 2 features from the dataframe were selected and used to construct the new feature? Or is there another reason?

Thanks in advance! I will continue reading the MDR paper I found on pubmed in the meanwhile.

Fixed older tests by adding hard-coded result arrays for testing fit and fit_transform.

Added custom metrics by importing accuracy_score, zero_one_loss from sklearn.

An error occurs when the class labels are not 0 and 1. When counting the number of cases and controls for each cell in a grid,

MDR code uses the following code (line 76-78):

       for row_i in range(features.shape[0]):
            feature_instance = tuple(features[row_i])
            self.class_count_matrix[feature_instance][classes[row_i]] += 1

classes is an array of y values passed as a parameter. Think of class_count_matrix as a (# of possible feature_instances) by (# of classes). Then since MDR takes in only binary data, # of classes is always 2 and therefore appropriate indices would be 0 and 1 for the dimension. But if the class labels are 0 and 2 not 0 and 1, then the program will try to index the class_count_matrix as class_count_matrix[(tuple of a feature_instance)][2], which is out of bounds.

Error message:

  File "<ipython-input-180-e1715a88facf>", line 10, in <module>
    mdr.fit(X_train, y_train)

  File "C:\Users\Hayley Son\Anaconda3\lib\site-packages\mdr\mdr.py", line 78, in fit
    self.class_count_matrix[feature_instance][classes[row_i]] += 1

IndexError: index 2 is out of bounds for axis 0 with size 2

Unit tests written for init, fit, transform, fit_transform using hard-coded dataset. Have not written test for scoring and have not implemented optional scoring metric from sklearn

Assumption(s): Labels are only binary

Implemented:

fit(self, features, classes): simply build a dictionary that maps each instance of the feature vector to a tuple. The tuple keeps count of how many times a particular label value appears with that instance of feature vector. Key: tuple of feature values - Value: tuple of label frequency/label counts

transform(self, features): After the dictionary is completed, combine each instance of feature vector above into one corresponding label that has the frequency ratio greater than its standard default ratio.

score(self, features, classes): Compare the new combined feature vector with its corresponding class labels, and count the times the two match. Output the average accuracy by averaging the match count over the length of the new feature vector / classes vector.

Implementation is tested in main() by training MDR on the training set and getting accuracy_score on the test set.

genetic_data = pd.read_csv('https://github.com/EpistasisLab/scikit-mdr/blob/master/data/GAMETES_Epistasis_2-Way_20atts_0.4H_EDM-1_1.tsv.gz', sep='\t', compression='gzip')

The above command is not working, error says : "BadGzipFile: Not a gzipped file (b'\n\n')"

It seems that the utilities in mdr.utils is designed for this purpose but there is no documentation about how to use them. I have a quick look into those codes and made the demo for calculating scores for n-way combinations and I think it maybe a way to finding feature combinations with epistatic effect. Please let me know if it is the correct way.

from mdr import MDRClassifier
import pandas as pd
from mdr.utils import n_way_models
import operator

genetic_data = pd.read_csv('https://github.com/EpistasisLab/scikit-mdr/raw/development/data/GAMETES_Epistasis_2-Way_20atts_0.4H_EDM-1_1.tsv.gz', sep='\t', compression='gzip')

features = genetic_data.drop('class', axis=1).values
labels = genetic_data['class'].values
feature_names = list(genetic_data.columns)

my_mdr = MDRClassifier()
my_mdr.fit(features, labels)
print("Score for using all features", my_mdr.score(features, labels))

#n: list (default: [2])
#The maximum size(s) of the MDR model to generate.
#e.g., if n == [3], all 3-way models will be generated.
n = [2]
mdr_score_list = []
#  Note that this function performs an exhaustive search through all feature combinations and can be computationally expensive.
for _, mdr_model_score, model_features in n_way_models(my_mdr, features, labels, n=n, feature_names=feature_names):
    mdr_score_list.append((model_features, mdr_model_score))
mdr_score_list.sort(key=operator.itemgetter(1), reverse=True)
print("The combination with highest score:", mdr_score_list[0])

Exported output:

Score for using all features 0.998125
The combination with highest score: (['P1', 'P2'], 0.793125)

Hi, I have a ChIP-seq style dataset of RPKM values that I want to reduce from 200x200000 into 200x1000, so that I only end up with 1000 variables at the end of the MDR process, for my 200 records.

What would be the recommended way to use scikit-mdr for this task?

I tried to install scikit-mdr on an Ubuntu 14.04 Linux via pip install but got this error below. To make sure it wasn't a versions issue with scikit-learn, I did a sudo pip install -U scikit-learn, which completed successfully, then tried to load MDR on a python console. See below.

Any ideas?

Successfully installed scikit-learn
Cleaning up...
avilella@ubuntu14:~$ 
avilella@ubuntu14:~$ python
Python 2.7.6 (default, Oct 26 2016, 20:30:19) 
[GCC 4.8.4] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from mdr import MDR
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.7/dist-packages/mdr/__init__.py", line 23, in <module>
    from .mdr_ensemble import MDREnsemble
  File "/usr/local/lib/python2.7/dist-packages/mdr/mdr_ensemble.py", line 26, in <module>
    from sklearn.ensemble import BaggingClassifier
ImportError: cannot import name BaggingClassifier

Instead of keeping all of the modules in the same utils.py file, break them out into separate submodules. This will help prevent situations where, for example, a user imports matplotlib when they're only using the n_way_models function (that doesn't use matplotlib).