preprocess_data.py

"""
Adapted from https://www.kaggle.com/myonin/music-recommendation-random-forest-xgboost
"""
#from sklearn import cross_validation, grid_search, metrics, ensemble
import xgboost as xgb
import numpy as np
import pandas as pd

PATH = "data/"
PATH_OUT = "clean_data/"

# Load data
df = pd.read_csv(f'{PATH}/train.csv')

# 1% sample of items
# df = df.sample(frac=0.01)

# Load and join songs data
songs = pd.read_csv(f'{PATH}/songs.csv')
df = pd.merge(df, songs, on='song_id', how='left')
del songs

# Load and join songs data
members = pd.read_csv(f'{PATH}/members.csv')
df = pd.merge(df, members, on='msno', how='left')
del members

# Replace NA
for i in df.select_dtypes(include=['object']).columns:
    df[i][df[i].isnull()] = 'unknown'
df = df.fillna(value=0)

# Create Dates

# registration_init_time
df.registration_init_time = pd.to_datetime(df.registration_init_time, format='%Y%m%d', errors='ignore')
df['registration_init_time_year'] = df['registration_init_time'].dt.year
df['registration_init_time_month'] = df['registration_init_time'].dt.month
df['registration_init_time_day'] = df['registration_init_time'].dt.day

# expiration_date
df.expiration_date = pd.to_datetime(df.expiration_date,  format='%Y%m%d', errors='ignore')
df['expiration_date_year'] = df['expiration_date'].dt.year
df['expiration_date_month'] = df['expiration_date'].dt.month
df['expiration_date_day'] = df['expiration_date'].dt.day

# Dates to categoty
df['registration_init_time'] = df['registration_init_time'].astype('category')
df['expiration_date'] = df['expiration_date'].astype('category')

# Object data to category
for col in df.select_dtypes(include=['object']).columns:
    df[col] = df[col].astype('category')

# Encoding categorical features
for col in df.select_dtypes(include=['category']).columns:
    df[col] = df[col].cat.codes

# Drop columns
df = df.drop(['expiration_date', 'lyricist'], 1)

df.to_csv(f"{PATH_OUT}/output.csv")