Skip to content

A python library for repurposing traditional classification-based resampling techniques for regression tasks

License

Notifications You must be signed in to change notification settings

atif-hassan/Regression_ReSampling

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

69 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

forthebadge made-with-python ForTheBadge built-with-love

PyPI version shields.io Downloads Maintenance

Regression ReSampling

A python library for repurposing traditional classification-based resampling (undersampling and/or oversampling) techniques for regression tasks. Currently supports all resampling techniques present in imblearn

Why does this exist?

While we were working on a regression task, we realized that the target variable was skewed, i.e., most samples were present in a particular range. One can easily solve the skew problem for classification tasks via a slew of resampling techniques (either under or over sampling) but this luxury is unavailable for regression tasks. We therefore decided to create an interface that can repurpose all resampling techniques for classification problems to regression problems!

How to install?

pip install reg_resampler

Functions and parameters

# This returns a numpy list of classes for each corresponding sample. It also automatically merges classes when required
fit(X, target, bins=3, min_n_samples=6, balanced_binning=False, verbose=2)
  • X - Either a pandas dataframe or numpy matrix. Complete data to be resampled.
  • target - Either string (for pandas) or index (for numpy). The target variable to be resampled.
  • bins=3 - The number of classes that the user wants to generate. (Default: 3)
  • min_n_samples=6 - Minimum number of samples in each bin. Bins having less than this value will be merged with the closest bin. Has to be more than neighbours in imblearn. (Default: 6)
  • balanced_binning=False - Decides whether samples are to be distributed roughly equally across all classes. (Default: False)
  • verbose=2 - 0 will disable print by package, 1 will print info about class mergers and 2 will also print class distributions.
# Performs resampling and returns the resampled dataframe/numpy matrices in the form of data and target variable.
resample(sampler_obj, trainX, trainY)
  • sampler_obj - Your favourite resampling algorithm's object (currently supports imblearn)
  • trainX - Either a pandas dataframe or numpy matrix. Data to be resampled. Also, contains the target variable
  • trainY - Numpy array of psuedo classes obtained from fit function.

Important Note

All functions return the same data type as provided in input.

How to import?

from reg_resampler import resampler

Usage

# Initialize the resampler object
rs = resampler()

# You might recieve info about class merger for low sample classes
# Generate classes
Y_classes = rs.fit(df_train, target=target, bins=num_bins)
# Create the actual target variable
Y = df_train[target]

# Create a smote (over-sampling) object from imblearn
smote = SMOTE(random_state=27)

# Now resample
final_X, final_Y = rs.resample(smote, df_train, Y_classes)

Tutorials

You can find further tutorials on how to use this library for cross-validation

Future Ideas

  • Support for more resampling techniques

Feature Request

Drop us an email at [email protected] or [email protected] if you want any particular feature

About

A python library for repurposing traditional classification-based resampling techniques for regression tasks

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published