Questions about sample code?

[megancooper]

Hello, I am new to python and machine learning but need to use the library for a project. I read the website and the sample code but am still confused on how I can retrieve the features that have been (selected?) by each of the Relief algorithms.

Apologies if the site goes over this, but I didn't see any information on this. I had a couple questions:

1. How do we get back the features selected by each algorithm?
2. The sample code below for the ReliefF algorithm prints a number at the end of running the code, is this number relevant to feature selection?

import pandas as pd
import numpy as np
from sklearn.pipeline import make_pipeline
from skrebate import ReliefF
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score

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

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

clf = make_pipeline(ReliefF(n_features_to_select=2, n_neighbors=100),
                    RandomForestClassifier(n_estimators=100))

print(np.mean(cross_val_score(clf, features, labels)))
>>> 0.795

Thanks for any help, I've been trying to figure out this code using the internet for a couple weeks now but have not really gotten anywhere

[ryanurbs]

Hi Megan, go to the user guide link on github: https://epistasislab.github.io/scikit-rebate/using/ and scroll down to acquiring feature importance scores. Ryan Using skrebate - scikit-rebate - GitHub Pages<https://epistasislab.github.io/scikit-rebate/using/> epistasislab.github.io We have designed the Relief algorithms to be integrated directly into scikit-learn machine learning workflows. Below, we provide code samples showing how the various Relief algorithms can be used as feature selection methods in scikit-learn pipelines. Ryan J. Urbanowicz, Ph.D. Assistant Professor of Informatics Perelman School of Medicine University of Pennsylvania ---------------------------------------------- 629 Blockley Hall 423 Guardian Drive University of Pennsylvania Philadelphia, Pennsylvania 19104 W. Phone: 215-746-4225 C. Phone: 802-299-9461 Web: www.ryanurbanowicz.com<http://www.ryanurbanowicz.com/> Twitter: www.twitter.com/DocUrbs<http://www.twitter.com/DocUrbs> ----------------------------------------------

________________________________ From: Megan <notifications@github.com> Sent: Monday, January 21, 2019 12:34:07 PM To: EpistasisLab/scikit-rebate Cc: Subscribed Subject: [External] [EpistasisLab/scikit-rebate] Questions about sample code? (#57) Hello, I am new to python and machine learning but need to use the library for a project. I read the website and the sample code but am still confused on how I can retrieve the features that have been (selected?) by each of the Relief algorithms. Apologies if the site goes over this, but I didn't see any information on this. I had a couple questions: 1. How do we get back the features selected by each algorithm? 2. The sample code below for the ReliefF algorithm prints a number at the end of running the code, is this number relevant to feature selection? import pandas as pd import numpy as np from sklearn.pipeline import make_pipeline from skrebate import ReliefF from sklearn.ensemble import RandomForestClassifier from sklearn.model_selection import cross_val_score genetic_data = pd.read_csv('https://github.com/EpistasisLab/scikit-rebate/raw/master/data/' 'GAMETES_Epistasis_2-Way_20atts_0.4H_EDM-1_1.tsv.gz', sep='\t', compression='gzip') features, labels = genetic_data.drop('class', axis=1).values, genetic_data['class'].values clf = make_pipeline(ReliefF(n_features_to_select=2, n_neighbors=100), RandomForestClassifier(n_estimators=100)) print(np.mean(cross_val_score(clf, features, labels)))

>> 0.795

Thanks for any help, I've been trying to figure out this code using the internet for a couple weeks now but have not really gotten anywhere — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#57>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ANWn0dKHEcw4vBvqVcP7drWMjBSRptGPks5vFfoPgaJpZM4aLQw->.

[megancooper]

Hi Ryan,

Thanks for the response, this helps alot!

Just to clarify, if I received the following result I would pick the highest scoring (most closest to 1?) features from my dataset?

>>>N0    -0.0000166666666667
>>>N1    -0.006175
>>>N2    -0.0079
>>>N3    -0.006275
>>>N4    -0.00684166666667
>>>N5    -0.0104416666667
>>>N6    -0.010275
>>>N7    -0.00785
>>>N8    -0.00824166666667
>>>N9    -0.00515
>>>N10   -0.000216666666667
>>>N11   -0.0039
>>>N12   -0.00291666666667
>>>N13   -0.00345833333333
>>>N14   -0.00324166666667
>>>N15   -0.00886666666667
>>>N16   -0.00611666666667
>>>N17   -0.007325
>>>P1    0.108966666667
>>>P2    0.111

[megancooper]

Actually my bad, was listed on the link as well:

To sort features by decreasing score along with their names, and simultaneously indicate which features have been assigned a token TuRF feature score (since they were removed from consideration at some point) then add the following...

scored\_features = len(fs.top\_features_)
sorted_names = sorted(scoreDict, key=lambda x: scoreDict[x], reverse=True)
n = 1
for k in sorted\_names:
    if n < scored\_features +1 :
        print(k, '\t', scoreDict[k],'\t',n) 
    else:
        print(k, '\t', scoreDict[k],'\t','*') 
    n += 1

>>>P1    0.20529375      1
>>>P2    0.17374375      2
>>>N0    -0.00103125     3
>>>N10   -0.00118125     4
>>>N13   -0.0086125      5
>>>N1    -0.0107515625   *
>>>N14   -0.0107515625   *
>>>N16   -0.0107515625   *
>>>N8    -0.0107515625   *
>>>N12   -0.0107515625   *
>>>N3    -0.012890625    *
>>>N2    -0.012890625    *
>>>N7    -0.012890625    *
>>>N17   -0.012890625    *
>>>N5    -0.012890625    *
>>>N15   -0.012890625    *
>>>N11   -0.012890625    *
>>>N4    -0.012890625    *
>>>N9    -0.012890625    *
>>>N6    -0.012890625    *

[megancooper]

@ryanurbs I had another question about allowable datatypes. Are strings not supported by this library? I noticed most of the sample data in this repo contains numbers for each feature and no strings. I am currently trying to use data that has strings, and I receive the following error:

TypeError: unsupported operand type(s) for /: 'str' and 'int'

My data looks something like this:

feature1	feature2	feature3	feature4
red	on	large	open
blue	off	small	open

My current code:

feature_pairs = pd.DataFrame(feature_value_pairs)

# Separate the features, from the label(s) (bug name(s))
features, labels = feature_pairs.drop('class', axis=1).values, feature_pairs['class'].values

# Make sure to compute the feature importance scores from only your training set
X_train, X_test, y_train, y_test = train_test_split(features, labels)

fs = ReliefF()
fs.fit(X_train, y_train) # This is where the TypeError occurs

[ryanurbs]

It was supposed to have been set up to handle strings as well, but I'll have to take a closer look, not sure when I will be able to get to that. In the meantime I'd suggest encoding your variables as integers to avoid the error. Thanks Ryan Get Outlook for Android<https://aka.ms/ghei36>

…

________________________________ From: Megan <notifications@github.com> Sent: Wednesday, January 23, 2019 3:40:31 PM To: EpistasisLab/scikit-rebate Cc: Ryan Urbanowicz; Mention Subject: [External] Re: [EpistasisLab/scikit-rebate] Questions about sample code? (#57) @ryanurbs<https://github.com/ryanurbs> I had another question about allowable datatypes. Are strings not supported by this library? I noticed most of the sample data in this repo contains numbers for each feature and no strings. I am currently trying to use data that has strings, and I receive the following error: TypeError: unsupported operand type(s) for /: 'str' and 'int' My data looks something like this: feature1 feature2 feature3 feature4 red on large open blue off small open My current code: feature_pairs = pd.DataFrame(feature_value_pairs) # Separate the features, from the label(s) (bug name(s)) features, labels = feature_pairs.drop('class', axis=1).values, feature_pairs['class'].values # Make sure to compute the feature importance scores from only your training set X_train, X_test, y_train, y_test = train_test_split(features, labels) fs = ReliefF() fs.fit(X_train, y_train) # This is where the TypeError occurs — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#57 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ANWn0RwCJc4Q170pYo6fpyYFxFR23LzIks5vGMi_gaJpZM4aLQw->.

[megancooper]

Apologies for the many questions, but I tried encoding my data and I still have the same error. It is happening everytime on line 140 of relieff.py:

self._labels_std = np.std(self._y, ddof=1)

Here is the full traceback:

Traceback (most recent call last):
line 36, in <module> fs.fit(X_train, y_train)
line 140, in fit self._labels_std = np.std(self._y, ddof=1)
line 3038, in std **kwargs)
line 140, in _std keepdims=keepdims)
line 110, in _var arrmean, rcount, out=arrmean, casting='unsafe', subok=False)
TypeError: unsupported operand type(s) for /: 'str' and 'int'

I checked my labels array and it seems to be formatted correctly. Are there any rules surrounding how these should be formatted? I am directly creating a DataFrame instead of using a tsv file like some of the examples show. Perhaps this is related?

Could it have something to do with the data types for my data frame's columns? I noticed it is a numpy error that is happening when np.std() is called, and that each of my columns are of the type object.

[txsing]

Hello, I am new to python and machine learning but need to use the library for a project. I read the website and the sample code but am still confused on how I can retrieve the features that have been (selected?) by each of the Relief algorithms.

Apologies if the site goes over this, but I didn't see any information on this. I had a couple questions:

1. How do we get back the features selected by each algorithm?
2. The sample code below for the ReliefF algorithm prints a number at the end of running the code, is this number relevant to feature selection?

import pandas as pd
import numpy as np
from sklearn.pipeline import make_pipeline
from skrebate import ReliefF
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score

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

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

clf = make_pipeline(ReliefF(n_features_to_select=2, n_neighbors=100),
                    RandomForestClassifier(n_estimators=100))

print(np.mean(cross_val_score(clf, features, labels)))
>>> 0.795

Thanks for any help, I've been trying to figure out this code using the internet for a couple weeks now but have not really gotten anywhere

I met the same problem, it seems a little bit difficult to find clear instructions on how to get ReliefF object from the pipeline object and to get to know the final selected features. I kept getting 'AttributeError: 'ReliefF' object has no attribute 'feature_importances_' error prompt by calling print(clf['relieff'].feature_importances_)

It will be great if the developer could give a simpler version of the example code showing the intermediate steps without using pipeline.