Scikit Pierre

Docs | Paper

Overview

Scikit-Pierre is a Scientific ToolKit for Post-processing Recommendations. Written in Python, the library is focused on the researcher's perspective, providing an experimental environment based on the state-of-the-art. Pierre works through a pass data and returns results. It receives a set of data and from it generates a recommendation list as return.

Inspired by scikit-learn and Surpise, Scikit-Pierre is composed of post-processing recommendation methods that are provided as api-classes.

• First of all, instance one class and pass the data.
• Second, set up the experiment with context functions.
• Third, fit the context and receive the recommendation list.

Attention

Pierre is a free library and has not direct grant, i.e., it is made with love and science. By this, please if you find some error/issue, remember that we have no grant support and demonstrate some kind of education in contact us. We will investigate the error and try to solve.

Built by

The original code draft is made by Diego Corrêa da Silva as part of his master dissertation and PhD thesis at Federal University of Bahia❤️.

Development Status

Pierre was designed with the following contexts in mind:

• Version 0.0.1 comprehends the context of Calibrated Recommendations.
• Version 0.0.2 supports Calibrated Popularity, Time and Entropy distribution derivation.
• Version 0.0.3 Incoming

Install and Config

With python and pip do:

$ pip install scikit-pierre

For the latest version, you can also clone the repo and build the source:

$ git clone https://github.com/scikit-pierre/scikit-pierre
$ cd scikit-pierre
$ conda create --name scikit-pierre 
$ conda activate scikit-pierre
$ conda install setuptools numpy pandas cython
$ python setup.py sdist bdist_wheel
$ pip install dist/scikit_pierre-0.0.1-py3-none-any.whl

Windows users might prefer using conda. We do not support Windows issues, we will try, but...

Usage example

from scikit_pierre.tradeoff.calibration import LinearCalibration
import pandas as pd

# Load the users' preference set as a Pandas DataFrame instance.
# It is expected 3 columns: [USER_ID, ITEM_ID, TRANSACTION_VALUE].
users_preference_dataframe = pd.read_csv('dataset/train.csv')

# Load the users' candidate items set as a Pandas DataFrame instance.
# It is expected 3 columns: [USER_ID, ITEM_ID, TRANSACTION_VALUE].
candidate_items_dataframe = pd.read_csv('dataset/candidate_items.csv')

# Load the items set as a Pandas DataFrame instance.
# It is expected 2 columns: [ITEM_ID, GENRES].
items_dataframe = pd.read_csv('dataset/items.csv')

# Create an instance with the basic data.
tradeoff_instance = LinearCalibration(
  users_preferences=users_preference_dataframe,
  candidate_items=candidate_items_dataframe,
  item_set=items_dataframe
)

# Configure the instance
tradeoff_instance.config(
  distribution_component='CWS',
  fairness_component='KL',
  relevance_component='SUM',
  tradeoff_weight_component='VAR',
  select_item_component='SURROGATE',
  list_size=10
)

# Execute the instance and get the recommendation list to all users.
recommendation_lists = tradeoff_instance.fit()
for user_recommendation in recommendation_lists[:2]:
    # print the 5 first in the dataframe
    print(user_recommendation.head(5))

Output:

   ITEM_ID  ORDER  USER_ID  TRANSACTION_VALUE
0       53      1        1           4.752746
1      905      2        1           4.632595
2     3365      3        1           4.534034
3     2565      4        1           4.499202
4     3077      5        1           4.426636
   ITEM_ID  ORDER  USER_ID  TRANSACTION_VALUE
0     2762      1        2           4.560991
1     1035      2        2           4.390060
2      953      3        2           4.366187
3      905      4        2           4.165182
4     3365      5        2           4.128045

Acknowledgments

We need to thank some import support to construct the Pierre:

• Student grant

  • The Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 88887.502736 / 2020–00. For provide the Diego's master degree grant;
  • The Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 88887.685243 / 2022-00. For provide the Diego's Ph.D. degree grant;

• Machine support

  • The Brazilian National Laboratory of Scientific Computing (LNCC/MCTI, Brazil ❤️ ), URL: http://sdumont.lncc.br. For provide the access to the Santos Dumont HPC;
  • The Center for Computational Biology and Biotechnological Information Management - NBCGI, with resources from FINEP/ MCT, CNPQ and FAPESB and from the State University of Santa Cruz - UESC.

People Contribution

• Diego Corrêa da Silva

References