skestimate 0.0.3

fit estimate utility

skestimate package

This package is used to fit and get various metrics for a multi class classifier. It is based on scikit-learn library and uses the methods from that library.

fit_est module inside the package has a class named Xest(self, estimator, data, target_label, ts).

• estimator: A piplined classifier that encodes all the categorical features.
• data: Raw dataset including the target label
• target_label: A string type representing the name of the target label
• ts: A number between 0 and 1 that specifies the test portion of the data.

Below is an example of how to use the class methods on the CoverType data set from UCI repository.

$ data = pd.read_csv("https://github.com/skhabiri/PredictiveModeling-CoverType-u2build/blob/master/data/train.csv?raw=true")
rfc = make_pipeline(
    RandomForestClassifier(bootstrap=True, ccp_alpha=0.0, class_weight=None,
                           criterion='entropy', max_depth=14, max_features=20,
                           max_leaf_nodes=None, max_samples=None,
                           min_impurity_decrease=0.0, min_impurity_split=None,
                           min_samples_leaf=2, min_samples_split=10,
                           min_weight_fraction_leaf=0.0, n_estimators=10, n_jobs=-1,
                           oob_score=False, random_state=42, verbose=0,
                           warm_start=False)
                    )
xest = skestimate.Xest(rfc, data, "Cover_Type", 0.2)

For local testing we can use the example() function in fit_est.py.

>>> import skestimate	
>>> myest = skestimate.example()
>>> myest.xskew(0.9)

Available methods associated with Xest class:

• xunique(): Reports counts of unique values in each column of data

• xskew(imb=0.99):
  Returns a pandas Series of the sorted column with skewness more than imb. imb is between 0 and 1

• xfit():
  Fits the pipeline estimator and returns fitted estimator, training score, and test score

• xscore(fit=True):
  Calculates accuracy, recall and precision of a classifier and plots the confusion matrix

Docker Container

An axample of how to test the sketimate package in a debian OS through docker image. Here are the steps:

• Create the Dockerfile in project (repository) directory. The specific instructions in Dockerfile includes:
  • Based in”debian” image in dockerhub
  • bash shell
  • Python3 and pip3 installed
  • numpy, pandas, scikit-learn, and matplotlib all installed
  • skestimate package installed
• Build the image on local machine with docker build . -t skestimate_di
• Very the existence of the new package with docker image ls
• Create and enter a fresh container with `docker run -it skestimate_di’
• Test the package with:
  • python3
  • import skestimate
  • est = skestimate.example()
  • est.xscore(fit=True)
  • exit() #exit the python repl
• Exit the container with exit

Dockerfile

## Image to base ours on
## docker run debian
FROM debian

## Environment variables to configure things
## Setting shell so pipenv shell works
ENV SHELL=/bin/bash

## Update and install dependencies
## Running them in pipeline. \ is for continue into a new line
## update, updates the list of packages available for install
## and install all the security packages to have an up to date OS.
RUN apt update && \
  ## Yes to possible questions as it is not running interactively
  ## upgrade potentially installs new versions of outdated packages
  apt upgrade -y && \
  ## install python3 and pip3 at the same time
  apt install python3-pip -y && \
  pip3 install pandas numpy scikit-learn matplotlib && \
  pip3 install skestimate

## In command prompt:
## docker build . -t skestimate_di
## docker run -it skestimate_di

Unit Testing

You can test the individual methods in command line with python test_skestimate.py --verbose