Refactor application foreground features
Meng Li
parent
d1f641c596
commit
9048c06fc4
|
|
@ -0,0 +1,64 @@
|
|||
import pandas as pd
|
||||
import itertools
|
||||
from scipy.stats import entropy
|
||||
|
||||
|
||||
def compute_features(filtered_data, apps_type, requested_features, apps_features, day_segment):
|
||||
if "timeoffirstuse" in requested_features:
|
||||
time_first_event = filtered_data.sort_values(by="timestamp", ascending=True).drop_duplicates(subset="local_date", keep="first").set_index("local_date")
|
||||
apps_features["apps_" + day_segment + "_timeoffirstuse" + apps_type] = time_first_event["local_hour"] * 60 + time_first_event["local_minute"]
|
||||
if "timeoflastuse" in requested_features:
|
||||
time_last_event = filtered_data.sort_values(by="timestamp", ascending=False).drop_duplicates(subset="local_date", keep="first").set_index("local_date")
|
||||
apps_features["apps_" + day_segment + "_timeoflastuse" + apps_type] = time_last_event["local_hour"] * 60 + time_last_event["local_minute"]
|
||||
if "frequencyentropy" in requested_features:
|
||||
apps_with_count = filtered_data.groupby(["local_date","application_name"]).count().sort_values(by="timestamp", ascending=False).reset_index()
|
||||
apps_features["apps_" + day_segment + "_frequencyentropy" + apps_type] = apps_with_count.groupby("local_date")["timestamp"].agg(entropy)
|
||||
if "count" in requested_features:
|
||||
apps_features["apps_" + day_segment + "_count" + apps_type] = filtered_data.groupby(["local_date"]).count()["timestamp"]
|
||||
apps_features.fillna(value={"apps_" + day_segment + "_count" + apps_type: 0}, inplace=True)
|
||||
return apps_features
|
||||
|
||||
|
||||
def base_applications_foreground_features(apps_data, day_segment, requested_features, params):
|
||||
multiple_categories_with_genres = params["multiple_categories_with_genres"]
|
||||
single_categories = params["single_categories"]
|
||||
multiple_categories = params["multiple_categories"]
|
||||
apps = params["apps"]
|
||||
|
||||
# deep copy the apps_data for the top1global computation
|
||||
apps_data_global = apps_data.copy()
|
||||
|
||||
if apps_data.empty:
|
||||
apps_features = pd.DataFrame(columns=["local_date"] + ["apps_" + day_segment + "_" + x for x in ["".join(feature) for feature in itertools.product(requested_features, single_categories + multiple_categories + apps)]])
|
||||
else:
|
||||
if day_segment != "daily":
|
||||
apps_data =apps_data[apps_data["local_day_segment"] == day_segment]
|
||||
|
||||
if not apps_data.empty:
|
||||
apps_features = pd.DataFrame()
|
||||
# single category
|
||||
for sc in single_categories:
|
||||
if sc == "all":
|
||||
apps_features = compute_features(apps_data, "all", requested_features, apps_features, day_segment)
|
||||
else:
|
||||
filtered_data = apps_data[apps_data["genre"].isin([sc])]
|
||||
apps_features = compute_features(filtered_data, sc, requested_features, apps_features, day_segment)
|
||||
# multiple category
|
||||
for mc in multiple_categories:
|
||||
filtered_data = apps_data[apps_data["genre"].isin(multiple_categories_with_genres[mc])]
|
||||
apps_features = compute_features(filtered_data, mc, requested_features, apps_features, day_segment)
|
||||
# single apps
|
||||
for app in apps:
|
||||
col_name = app
|
||||
if app == "top1global":
|
||||
# get the most used app
|
||||
apps_with_count = apps_data_global.groupby(["local_date","package_name"]).count().sort_values(by="timestamp", ascending=False).reset_index()
|
||||
app = apps_with_count.iloc[0]["package_name"]
|
||||
col_name = "top1global"
|
||||
|
||||
filtered_data = apps_data[apps_data["package_name"].isin([app])]
|
||||
apps_features = compute_features(filtered_data, col_name, requested_features, apps_features, day_segment)
|
||||
|
||||
apps_features = apps_features.reset_index()
|
||||
|
||||
return apps_features
|
||||
|
|
@ -1,24 +1,5 @@
|
|||
import pandas as pd
|
||||
import numpy as np
|
||||
import itertools
|
||||
from scipy.stats import entropy
|
||||
|
||||
|
||||
def compute_features(filtered_data, apps_type, requested_features, apps_features):
|
||||
if "timeoffirstuse" in requested_features:
|
||||
time_first_event = filtered_data.sort_values(by="timestamp", ascending=True).drop_duplicates(subset="local_date", keep="first").set_index("local_date")
|
||||
apps_features["apps_" + day_segment + "_timeoffirstuse" + apps_type] = time_first_event["local_hour"] * 60 + time_first_event["local_minute"]
|
||||
if "timeoflastuse" in requested_features:
|
||||
time_last_event = filtered_data.sort_values(by="timestamp", ascending=False).drop_duplicates(subset="local_date", keep="first").set_index("local_date")
|
||||
apps_features["apps_" + day_segment + "_timeoflastuse" + apps_type] = time_last_event["local_hour"] * 60 + time_last_event["local_minute"]
|
||||
if "frequencyentropy" in requested_features:
|
||||
apps_with_count = filtered_data.groupby(["local_date","application_name"]).count().sort_values(by="timestamp", ascending=False).reset_index()
|
||||
apps_features["apps_" + day_segment + "_frequencyentropy" + apps_type] = apps_with_count.groupby("local_date")["timestamp"].agg(entropy)
|
||||
if "count" in requested_features:
|
||||
apps_features["apps_" + day_segment + "_count" + apps_type] = filtered_data.groupby(["local_date"]).count()["timestamp"]
|
||||
apps_features.fillna(value={"apps_" + day_segment + "_count" + apps_type: 0}, inplace=True)
|
||||
return apps_features
|
||||
|
||||
from applications_foreground.applications_foreground_base import base_applications_foreground_features
|
||||
|
||||
apps_data = pd.read_csv(snakemake.input[0], parse_dates=["local_date_time", "local_date"], encoding="ISO-8859-1")
|
||||
day_segment = snakemake.params["day_segment"]
|
||||
|
|
@ -27,11 +8,19 @@ multiple_categories_with_genres = snakemake.params["multiple_categories"]
|
|||
single_apps = snakemake.params["single_apps"]
|
||||
excluded_categories = snakemake.params["excluded_categories"]
|
||||
excluded_apps = snakemake.params["excluded_apps"]
|
||||
features = snakemake.params["features"]
|
||||
requested_features = snakemake.params["features"]
|
||||
apps_features = pd.DataFrame(columns=["local_date"])
|
||||
|
||||
single_categories = list(set(single_categories) - set(excluded_categories))
|
||||
multiple_categories = list(multiple_categories_with_genres.keys() - set(excluded_categories))
|
||||
apps = list(set(single_apps) - set(excluded_apps))
|
||||
type_count = len(single_categories) + len(multiple_categories) + len(apps)
|
||||
|
||||
params = {}
|
||||
params["multiple_categories_with_genres"] = multiple_categories_with_genres
|
||||
params["single_categories"] = single_categories
|
||||
params["multiple_categories"] = multiple_categories
|
||||
params["apps"] = apps
|
||||
|
||||
# exclude categories in the excluded_categories list
|
||||
if "system_apps" in excluded_categories:
|
||||
|
|
@ -40,38 +29,8 @@ apps_data = apps_data[~apps_data["genre"].isin(excluded_categories)]
|
|||
# exclude apps in the excluded_apps list
|
||||
apps_data = apps_data[~apps_data["application_name"].isin(excluded_apps)]
|
||||
|
||||
# deep copy the apps_data for the top1global computation
|
||||
apps_data_global = apps_data.copy()
|
||||
apps_features = apps_features.merge(base_applications_foreground_features(apps_data, day_segment, requested_features, params), on="local_date", how="outer")
|
||||
|
||||
apps_features = pd.DataFrame(columns=["local_date"] + ["apps_" + day_segment + "_" + x for x in ["".join(feature) for feature in itertools.product(features, single_categories + multiple_categories + apps)]])
|
||||
if not apps_data.empty:
|
||||
apps_features = pd.DataFrame()
|
||||
if day_segment != "daily":
|
||||
apps_data =apps_data[apps_data["local_day_segment"] == day_segment]
|
||||
|
||||
# single category
|
||||
for sc in single_categories:
|
||||
if sc == "all":
|
||||
apps_features = compute_features(apps_data, "all", features, apps_features)
|
||||
else:
|
||||
filtered_data = apps_data[apps_data["genre"].isin([sc])]
|
||||
apps_features = compute_features(filtered_data, sc, features, apps_features)
|
||||
# multiple category
|
||||
for mc in multiple_categories:
|
||||
filtered_data = apps_data[apps_data["genre"].isin(multiple_categories_with_genres[mc])]
|
||||
apps_features = compute_features(filtered_data, mc, features, apps_features)
|
||||
# single apps
|
||||
for app in apps:
|
||||
col_name = app
|
||||
if app == "top1global":
|
||||
# get the most used app
|
||||
apps_with_count = apps_data_global.groupby(["local_date","package_name"]).count().sort_values(by="timestamp", ascending=False).reset_index()
|
||||
app = apps_with_count.iloc[0]["package_name"]
|
||||
col_name = "top1global"
|
||||
|
||||
filtered_data = apps_data[apps_data["package_name"].isin([app])]
|
||||
apps_features = compute_features(filtered_data, col_name, features, apps_features)
|
||||
|
||||
apps_features = apps_features.reset_index()
|
||||
assert len(requested_features) * type_count + 1 == apps_features.shape[1], "The number of features in the output dataframe (=" + str(apps_features.shape[1]) + ") does not match the expected value (=" + str(len(requested_features)) + " + 1). Verify your application foreground feature extraction functions"
|
||||
|
||||
apps_features.to_csv(snakemake.output[0], index=False)
|
||||
|
|
|
|||
Loading…
Reference in New Issue