def preprocessing(months, user_ids=None, featurizer=None, features=None, prefilter=True, dfs=None):
'''
Takes a list of users_id, creates the relevant window from their first deposit date
and featurizes within it.
Args:
months: Number of months ahead the frame looks from the first deposit date
user_ids: List of integer user_ids to use in this sample
demo_df: The demographic info to pull the user's
featurizer: Optional featurizer object
features: Optional list of features to use, if none it'll use every feature in the featurizer
prefilter: Whether to apply prefilters such as activity threshold and rg-frame filtering
dfs: The information associated with the users
Returns:
X: ndarray of the the featurized rows
y: Labels associated with each row of X
user_ids: The user_ids associated with each row of X
'''
if not featurizer and not features:
print("Need at least one way to get featurizing context!")
raise ValueError
if not dfs:
demo_df, rg_df, gam_df = get_demo_df(), get_rg_df(), get_gam_df()
else:
demo_df, rg_df, gam_df = dfs
if not user_ids:
user_ids = list(demo_df.index)
days = months * 30
if prefilter:
print("Applying prefilters")
user_ids = prefilters(user_ids, months*30, demo_df, rg_df)
print(f"Constructing model with {months} months of information")
print(f"Features being used: {features}")
X, y = featurize(user_ids, gam_df, demo_df, featurizer=featurizer, features=features, month_window=months)
return X, y, user_ids
rgs: The label associated with each row
'''
print("Starting frame making")
if not featurizer:
featurizer = make_default_featurizer()
frames = [
make_frame(user_id, gam_df, demo_df, month_window)
for user_id in user_ids
]
rgs = [demo_df.loc[user_id, 'rg'] == 1 for user_id in user_ids]
return featurizer.vectorize(frames, features), rgs
def make_frame(user_id, gam_df, demo_df, month_window):
'''Featurizes a single user'''
mask = (gam_df['user_id'] == user_id)
user_daily = gam_df[mask]
first_deposit = demo_df.loc[user_id, 'first_deposit_date']
user_frame = sparse_to_ts(user_daily,
date_start=first_deposit,
window=30 * month_window)
return user_frame
if __name__ == '__main__':
demo_df = get_demo_df()
gam_df = get_gam_df()
rg_df = get_rg_df()
user_ids = list(demo_df.index)
print(len(user_ids))
return vect
def add_feature(self, prod_function, feat_name=None, args={}):
if not feat_name:
feat_name = prod_function.__name__
self.features[feat_name] = lambda x: prod_function(x, **args)
def delete_feature(self, feat_name):
del self.features[feat_name]
def get_feature_names(self):
return list(self.features.keys())
if __name__ == "__main__":
demo_df = pipeline.get_demo_df()
gam_df = pipeline.get_gam_df()
rg_info = pipeline.get_rg_df()
user_id = 3327778
featurizer = Featurizer()
featurizer.add_feature(total_hold)
featurizer.add_feature(max_hold)
featurizer.add_feature(weekly_hold)
featurizer.add_feature(weekly_rolling_hold)
mask = (gam_df['user_id'] == user_id)
user_daily = daily_gam_df[mask]
first_deposit = demo_df.loc[user_id, 'first_deposit_date']
user_frame = sparse_to_ts(user_daily, date_start=first_deposit, window=180)
features_to_use = ["total_hold", "max_hold", "weekly_hold"]
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib import rc, rcParams
from plot_helper import *
from sklearn.metrics import roc_curve
import pipeline
from pipeline_constants import *
rcParams.update({'figure.autolayout': True})
plt.style.use('ggplot')
demo_df = pipeline.get_demo_df() # Global vars weee
gam_df = pipeline.get_gam_df()
rg_info = pipeline.get_rg_df()
def background_plot(ax, user_id, gam_df, window=30 * 6):
'''Plots the introductory "Wow people lose a lot on this" graph'''
mask = (gam_df['user_id'] == user_id)
user_daily = gam_df[mask]
first_deposit = demo_df.loc[user_id, 'first_deposit_date']
user_frame = pipeline.sparse_to_ts(user_daily,
date_start=first_deposit,
window=window)
user_frame['cumul_hold'] = user_frame['hold'].cumsum()
ax.set_title(f'User #{user_id}', fontsize=28)
ax.set_xlabel("Date", fontsize=28)
ax.set_ylabel("Loss (Euros)", fontsize=28)
ax.tick_params(axis="y", labelsize=20)
ax.tick_params(axis="x", labelsize=20)
from pipeline import get_demo_df, get_gam_df, get_rg_df
from processing.features import SUMMARY_NAMES, DAILY_NAMES, WEEKLY_NAMES
from processing.preprocessing import preprocessing
from model import predict
if __name__ == '__main__':
print("Running on holdout!")
sleep(10)
HOLD_DEMO_PATH = 'data/holdout/demographic.csv'
HOLD_RG_PATH = 'data/holdout/rg_information.csv'
HOLD_GAM_PATH = 'data/holdout/gambling.csv'
hold_demo = get_demo_df(HOLD_DEMO_PATH)
hold_rg = get_rg_df(HOLD_RG_PATH)
hold_gam = get_gam_df(HOLD_GAM_PATH)
dfs = [hold_demo, hold_rg, hold_gam]
#model =
#features = S
X, y, user_ids = preprocessing(months=months, features=features, dfs=dfs)
predict(model, X, y, user_ids, store_name="holdout" store=True)