Copper

Wrapper around python data analysis packages such as pandas, scikit-learn and matplotlib to make data analysis on python easier.

Requirements

1. Python
2. pandas
3. scikit-learn
4. matplotlib
5. rpy2
6. tornado

Note: pandas is the only package that is required before installing copper, but is recommended to have all other packages installed too.

Note 2: The package is developed for Python 3 and Python 2 with a single code base. But the main target is Python 3 so is recommended since most packages already support Python 3.

Install

pip install copper

Features

1. Project structure for Data Analysis projects ala Project Template on R.
2. Dataset: Wrapper around pandas.DataFrame to introduce metadata
3. Data transformation templates
4. Custom matplotlib charts for exploration: histograms, scatterplots
5. Exploration via D3.js (very experimental)
6. More data imputation options via R (rpy2)
7. Rapid Machine Learning prototyping:
   • Easy to compare classifiers
   • Ensemble (bagging)

Project Structure

Copper uses a project structure based on Project Template (from R) to give structure to a Data Analysis project.

The suggested structure is:

data -> `project/data': All the data files, raw, cached, etc.

Is suggested to use /data/raw for raw files such as .csv files.

Copper by default loads data from the data folder. For example: data = copper.read_csv('catalog.csv') will load the project/data/catalog.csv file into a pandas.DataFrame using the pandas read_csv method and parameters.

As expected when saving files (copper.save(...) or copper.export(...)) copper saves the files on the data folder

source -> src: Python, iPython notebook files.

Following the intuition every file inside the source folder should do:

import copper
copper.project.path = '../'

For other suggested folders see: Project Template

For more info about this see the examples below.

Examples

Donors:

• Project structure and histograms

Loans:

• Automatic data transformation

Catalog:

• Custom transformation and basic machine learning

Kaggle Bulldozers:

• Post: Basic feature selection and join Datasets

For more information and more examples (but some are possible outdated) can see my blog: danielfrg.github.com