FeatureGarage

FeatureGarage: A fast and simple pandas.DataFrame r/w library for feature engineering

Installation

pip install -U git+https://github.com/69guitar1015/feature_garage

Basic Usage

from feature_garage import Garage
import pandas as pd

garage = Garage(dirpath='path/to/datadir')

df1 = pd.read_csv('path/to/csv')

# Get table list
print(garage.tables())

# Get feature list in a table
print(garage.columns(table='table'))

# Save all columns
garage.save(df1, table='df1')
# Save only specified columns
garage.save(df1, table='df1', columns=['f1', 'f2'])
# Overwrite existing columns
garage.save(df1, table='df1', overwrite=True)
# Make warning silent
garage.save(df1, table='df1', verbose=0)
# Garage.save reduces each column data size on default using np.min_scalar_dtype
garage.save(df1, table='df1', minimize_dtype=False)

# Load
df2 = garage.load(table='df2')

# Drop columns
garage.drop_columns(table='df1', columns=['f1', 'f2'])

How to save

FeatureGarage is just saving each column as array in npy format. table is just a directory and column(feature) is just a npy file.

categorical features may be saved as 'object'

Future Work

• Prallel r/w
• seemless loading for categorical features