A utility for taking discrete valued attributes and replacing with numerical values

data_util.py

import numpy as np
import pandas as pd

def clean_data(df, fill_method='ffill'):
	'''
	Takes in a pandas DataFrame and changes all string attributes and outputs
	to integers. Returns the DataFrame and the mapping
	fill_method can be:
	- 'ffill': forward fill
	- 'bfill': backward fill
	'''
	# Find all columns where data type is string
	text_cols = df.select_dtypes(include=['object']).columns.values
	text_cols = list(text_cols)
	mappings = dict()

	if text_cols:
		def apply_mapping(index):
			text = df.loc[index][colname]
			num = colname_mapping[text]
			df.loc[index, colname] = num

		for colname in text_cols:
			view = df[colname]
			distinct = view.unique()

			colname_mapping = {text : num for num, text in enumerate(distinct)}
			mappings[colname] = colname_mapping
			map(apply_mapping, df.index)


	# Replace True/False with 0 or 1
	bool_cols = df.select_dtypes(include=['bool']).columns.values
	bool_cols = list(bool_cols)

	if bool_cols:
		for colname in bool_cols:
			df[colname] = df[colname].astype(int)

	# # Fill missing values using forward filling
	df = df.fillna(method=fill_method)

	return (df, mappings)


def process_dataset(filename, outname):
	df = pd.read_csv(filename)
	colnames = list(df.columns)

	attributes = df[colnames[:-1]].copy()
	outputs = df[colnames[-1]]

	attributes, mappings = clean_data(attributes)
	#
	df = pd.concat([attributes, outputs], axis=1)
	print df

	df.to_csv(outname, index=False)


if __name__ == '__main__':
    filename = 'original.csv'  # Which file you're reading from
    outname = 'dataset.csv'  # Name of the file to write to (erases if already exists)
	process_dataset(filename, outname)

You'll need pandas and numpy

pip install pandas numpy