Before we can tap the potential that is buried in the app data, we need to join the apps to the device ids.
In [1]:
import numpy as np
import pandas as pd
import pickle
#path to data
DATA_PATH = "../../../input/"
Next, we load the data.
In [2]:
train = pd.read_csv("{0}gender_age_train.csv".format(DATA_PATH))
test = pd.read_csv("{0}gender_age_test.csv".format(DATA_PATH))
events = pd.read_csv("{0}events.csv".format(DATA_PATH))
apps = pd.read_csv("{0}app_events.csv".format(DATA_PATH))
apps['app_id'] = apps['app_id'].astype(str)
We merge first with the events data...
In [ ]:
train_test = pd.concat([train['device_id'], test['device_id']])
train_test_event = pd.merge(train_test.to_frame(), events[['device_id', 'event_id']], 'inner', on = 'device_id').drop_duplicates()
... and afterwards with the apps data.
In [ ]:
train_test_app = train_test_event.merge( apps,'left', on = 'event_id').drop('event_id', axis = 1).drop_duplicates()
Finally, we pickle the merged dataframe.
In [ ]:
pickle.dump(train_test_app, open('{0}device_apps_inner'.format(DATA_PATH), 'wb'))