230 lines
16 KiB
Python
230 lines
16 KiB
Python
import pandas as pd
|
|
from datetime import datetime
|
|
import itertools
|
|
|
|
def featuresFullNames(intraday_features_to_compute, sleep_levels_to_compute, sleep_types_to_compute, levels_include_all_groups):
|
|
|
|
features_fullname = ["local_segment"]
|
|
|
|
sleep_level_with_group = []
|
|
for sleep_level_group in sleep_levels_to_compute:
|
|
for sleep_level in sleep_levels_to_compute[sleep_level_group]:
|
|
sleep_level_with_group.append(sleep_level + sleep_level_group.lower())
|
|
|
|
if levels_include_all_groups:
|
|
features_fullname.extend([x[0] + x[1] + x[2] for x in itertools.product(intraday_features_to_compute["LEVELS_AND_TYPES"], sleep_level_with_group + ["all"], sleep_types_to_compute)])
|
|
else:
|
|
features_fullname.extend([x[0] + x[1] + x[2] for x in itertools.product(intraday_features_to_compute["LEVELS_AND_TYPES"], sleep_level_with_group, sleep_types_to_compute)])
|
|
if "ACROSS_LEVELS" in intraday_features_to_compute["RATIOS_SCOPE"]:
|
|
features_fullname.extend(["ratio" + x[0] + x[1] for x in itertools.product(intraday_features_to_compute["RATIOS_TYPE"], sleep_level_with_group)])
|
|
if "ACROSS_TYPES" in intraday_features_to_compute["RATIOS_SCOPE"] and "main" in sleep_types_to_compute:
|
|
features_fullname.extend(["ratio" + x + "main" for x in intraday_features_to_compute["RATIOS_TYPE"]])
|
|
if "WITHIN_LEVELS" in intraday_features_to_compute["RATIOS_SCOPE"] and "main" in sleep_types_to_compute:
|
|
features_fullname.extend(["ratio" + x[0] + "mainwithin" + x[1] for x in itertools.product(intraday_features_to_compute["RATIOS_TYPE"], sleep_level_with_group)])
|
|
if "WITHIN_TYPES" in intraday_features_to_compute["RATIOS_SCOPE"]:
|
|
features_fullname.extend(["ratio" + x[0] + x[1] + "within" + x[2] for x in itertools.product(intraday_features_to_compute["RATIOS_TYPE"], sleep_level_with_group, set(sleep_types_to_compute) & set(["main", "nap"]))])
|
|
|
|
return features_fullname
|
|
|
|
def mergeSleepEpisodes(sleep_data, cols_for_groupby):
|
|
|
|
sleep_episodes = pd.DataFrame(columns=["local_segment", "duration", "start_timestamp", "end_timestamp", "local_start_date_time", "local_end_date_time"])
|
|
|
|
if cols_for_groupby and (not sleep_data.empty):
|
|
sleep_data = sleep_data.groupby(by=cols_for_groupby, sort=False)
|
|
sleep_episodes = sleep_data[["duration"]].sum()
|
|
sleep_episodes["start_timestamp"] = sleep_data["start_timestamp"].first()
|
|
sleep_episodes["end_timestamp"] = sleep_data["end_timestamp"].last()
|
|
sleep_episodes["local_start_date_time"] = sleep_data["local_start_date_time"].first()
|
|
sleep_episodes["local_end_date_time"] = sleep_data["local_end_date_time"].last()
|
|
|
|
sleep_episodes.reset_index(inplace=True, drop=False)
|
|
|
|
return sleep_episodes
|
|
|
|
def statsFeatures(sleep_episodes, features, episode_type):
|
|
|
|
episode_features = pd.DataFrame(columns=[feature + episode_type for feature in features])
|
|
if sleep_episodes.empty:
|
|
return episode_features
|
|
|
|
if "countepisode" in features:
|
|
episode_features["countepisode" + episode_type] = sleep_episodes[["local_segment", "duration"]].groupby(["local_segment"])["duration"].count()
|
|
if "sumduration" in features:
|
|
episode_features["sumduration" + episode_type] = sleep_episodes[["local_segment", "duration"]].groupby(["local_segment"])["duration"].sum()
|
|
if "maxduration" in features:
|
|
episode_features["maxduration" + episode_type] = sleep_episodes[["local_segment", "duration"]].groupby(["local_segment"])["duration"].max()
|
|
if "minduration" in features:
|
|
episode_features["minduration" + episode_type] = sleep_episodes[["local_segment", "duration"]].groupby(["local_segment"])["duration"].min()
|
|
if "avgduration" in features:
|
|
episode_features["avgduration" + episode_type] = sleep_episodes[["local_segment", "duration"]].groupby(["local_segment"])["duration"].mean()
|
|
if "medianduration" in features:
|
|
episode_features["medianduration" + episode_type] = sleep_episodes[["local_segment", "duration"]].groupby(["local_segment"])["duration"].median()
|
|
if "stdduration" in features:
|
|
episode_features["stdduration" + episode_type] = sleep_episodes[["local_segment", "duration"]].groupby(["local_segment"])["duration"].std()
|
|
|
|
return episode_features
|
|
|
|
def allStatsFeatures(sleep_data, base_sleep_levels, base_sleep_types, features, sleep_intraday_features):
|
|
|
|
# For CLASSIC
|
|
for sleep_level, sleep_type in itertools.product(base_sleep_levels["CLASSIC"] + ["all"], base_sleep_types):
|
|
sleep_episodes_classic = sleep_data[sleep_data["type"] == "classic"]
|
|
sleep_episodes_classic = sleep_episodes_classic[sleep_episodes_classic["is_main_sleep"] == (1 if sleep_type == "main" else 0)] if sleep_type != "all" else sleep_episodes_classic
|
|
sleep_episodes_classic = sleep_episodes_classic[sleep_episodes_classic["level"] == sleep_level] if sleep_level != "all" else sleep_episodes_classic
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, statsFeatures(sleep_episodes_classic, features, sleep_level + "classic" + sleep_type)], axis=1)
|
|
|
|
# For STAGES
|
|
for sleep_level, sleep_type in itertools.product(base_sleep_levels["STAGES"] + ["all"], base_sleep_types):
|
|
sleep_episodes_stages = sleep_data[sleep_data["type"] == "stages"]
|
|
sleep_episodes_stages = sleep_episodes_stages[sleep_episodes_stages["is_main_sleep"] == (1 if sleep_type == "main" else 0)] if sleep_type != "all" else sleep_episodes_stages
|
|
sleep_episodes_stages = sleep_episodes_stages[sleep_episodes_stages["level"] == sleep_level] if sleep_level != "all" else sleep_episodes_stages
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, statsFeatures(sleep_episodes_stages, features, sleep_level + "stages" + sleep_type)], axis=1)
|
|
|
|
# For UNIFIED
|
|
for sleep_level, sleep_type in itertools.product(base_sleep_levels["UNIFIED"] + ["all"], base_sleep_types):
|
|
sleep_episodes_unified = sleep_data[sleep_data["is_main_sleep"] == (1 if sleep_type == "main" else 0)] if sleep_type != "all" else sleep_data
|
|
sleep_episodes_unified = sleep_episodes_unified[sleep_episodes_unified["unified_level"] == (0 if sleep_level == "awake" else 1)] if sleep_level != "all" else sleep_episodes_unified
|
|
sleep_episodes_unified = mergeSleepEpisodes(sleep_episodes_unified, ["local_segment", "unified_level_episode_id"])
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, statsFeatures(sleep_episodes_unified, features, sleep_level + "unified" + sleep_type)], axis=1)
|
|
|
|
# Ignore the levels (e.g. countepisode[all][main])
|
|
for sleep_type in base_sleep_types:
|
|
sleep_episodes_none = sleep_data[sleep_data["is_main_sleep"] == (1 if sleep_type == "main" else 0)] if sleep_type != "all" else sleep_data
|
|
sleep_episodes_none = mergeSleepEpisodes(sleep_episodes_none, ["local_segment", "type_episode_id"])
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, statsFeatures(sleep_episodes_none, features, "all" + sleep_type)], axis=1)
|
|
|
|
sleep_intraday_features.fillna(0, inplace=True)
|
|
|
|
return sleep_intraday_features
|
|
|
|
|
|
# Since all the stats features have been computed no matter they are requested or not,
|
|
# we can pick the related features to calculate the RATIOS features directly.
|
|
# Take ACROSS_LEVELS RATIOS features as an example:
|
|
# ratiocount[remstages] = countepisode[remstages][all] / countepisode[all][all]
|
|
def ratiosFeatures(sleep_intraday_features, ratios_types, ratios_scopes, sleep_levels, sleep_types):
|
|
|
|
# Put sleep_level_group and sleep_level together.
|
|
# For example:
|
|
# input (sleep_levels): {"CLASSIC": ["awake", "restless", "asleep"], "UNIFIED": ["awake", "asleep"]}
|
|
# output (sleep_level_with_group): [("classic", "awake"), ("classic", "restless"), ("classic", "asleep"), ("unified", "awake"), ("unified", "asleep")]
|
|
sleep_level_with_group = []
|
|
for sleep_level_group in sleep_levels:
|
|
for sleep_level in sleep_levels[sleep_level_group]:
|
|
sleep_level_with_group.append((sleep_level_group.lower(), sleep_level))
|
|
|
|
# ACROSS LEVELS
|
|
if "ACROSS_LEVELS" in ratios_scopes:
|
|
# Get the cross product of ratios_types and sleep_level_with_group.
|
|
# For example:
|
|
# input: ratios_types is ["count", "duration"], sleep_level_with_group is [("classic", "awake"), ("classic", "restless"), ("unified", "asleep")]
|
|
# output:
|
|
# 1) ratios_type: "count", sleep_levels_combined: ("classic", "awake")
|
|
# 2) ratios_type: "count", sleep_levels_combined: ("classic", "restless")
|
|
# 3) ratios_type: "count", sleep_levels_combined: ("unified", "asleep")
|
|
# 4) ratios_type: "duration", sleep_levels_combined: ("classic", "awake")
|
|
# 5) ratios_type: "duration", sleep_levels_combined: ("classic", "restless")
|
|
# 6) ratios_type: "duration", sleep_levels_combined: ("unified", "asleep")
|
|
for ratios_type, sleep_levels_combined in itertools.product(ratios_types, sleep_level_with_group):
|
|
sleep_level_group, sleep_level = sleep_levels_combined[0], sleep_levels_combined[1]
|
|
agg_func = "countepisode" if ratios_type == "count" else "sumduration"
|
|
across_levels = (sleep_intraday_features[agg_func + sleep_level + sleep_level_group + "all"] / sleep_intraday_features[agg_func + "all" + sleep_level_group + "all"]).to_frame().rename(columns={0: "ratio" + ratios_type + sleep_level + sleep_level_group})
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, across_levels], axis=1)
|
|
|
|
# ACROSS TYPES
|
|
if "ACROSS_TYPES" in ratios_scopes:
|
|
for ratios_type in ratios_types:
|
|
agg_func = "countepisode" if ratios_type == "count" else "sumduration"
|
|
# We do not provide the ratio for nap because is complementary.
|
|
across_types = (sleep_intraday_features[agg_func + "allmain"] / sleep_intraday_features[agg_func + "allall"]).to_frame().rename(columns={0: "ratio" + ratios_type + "main"})
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, across_types], axis=1)
|
|
|
|
# Get the cross product of ratios_types, sleep_level_with_group, and sleep_types.
|
|
# For example:
|
|
# input:
|
|
# ratios_types is ["count", "duration"]
|
|
# sleep_level_with_group is [("classic", "awake"), ("unified", "asleep")]
|
|
# sleep_types is ["main", "nap"]
|
|
# output:
|
|
# 1) ratios_type: "count", sleep_levels_combined: ("classic", "awake"), sleep_type: "main"
|
|
# 2) ratios_type: "count", sleep_levels_combined: ("classic", "awake"), sleep_type: "nap"
|
|
# 3) ratios_type: "count", sleep_levels_combined: ("unified", "asleep"), sleep_type: "main"
|
|
# 4) ratios_type: "count", sleep_levels_combined: ("unified", "asleep"), sleep_type: "nap"
|
|
# 5) ratios_type: "duration", sleep_levels_combined: ("classic", "awake"), sleep_type: "main"
|
|
# 6) ratios_type: "duration", sleep_levels_combined: ("classic", "awake"), sleep_type: "nap"
|
|
# 7) ratios_type: "duration", sleep_levels_combined: ("unified", "asleep"), sleep_type: "main"
|
|
# 8) ratios_type: "duration", sleep_levels_combined: ("unified", "asleep"), sleep_type: "nap"
|
|
for ratios_type, sleep_levels_combined, sleep_type in itertools.product(ratios_types, sleep_level_with_group, sleep_types):
|
|
|
|
# "all" sleep type will not be cosidered for any ratios features since it will be 1 all the time
|
|
if sleep_type == "all":
|
|
continue
|
|
|
|
sleep_level_group, sleep_level = sleep_levels_combined[0], sleep_levels_combined[1]
|
|
agg_func = "countepisode" if ratios_type == "count" else "sumduration"
|
|
|
|
# WITHIN LEVELS
|
|
if ("WITHIN_LEVELS" in ratios_scopes) and (sleep_type == "main"): # We do not provide the ratio for nap because is complementary.
|
|
within_levels = (sleep_intraday_features[agg_func + sleep_level + sleep_level_group + sleep_type] / sleep_intraday_features[agg_func + sleep_level + sleep_level_group + "all"]).to_frame().rename(columns={0: "ratio" + ratios_type + sleep_type + "within" + sleep_level + sleep_level_group})
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, within_levels], axis=1)
|
|
|
|
# WITHIN TYPES
|
|
if "WITHIN_TYPES" in ratios_scopes:
|
|
within_types = (sleep_intraday_features[agg_func + sleep_level + sleep_level_group + sleep_type] / sleep_intraday_features[agg_func + "all" + sleep_level_group + sleep_type]).to_frame().rename(columns={0: "ratio" + ratios_type + sleep_level + sleep_level_group + "within" + sleep_type})
|
|
sleep_intraday_features = pd.concat([sleep_intraday_features, within_types], axis=1)
|
|
|
|
return sleep_intraday_features
|
|
|
|
|
|
|
|
|
|
def rapids_features(sensor_data_files, time_segment, provider, filter_data_by_segment, *args, **kwargs):
|
|
|
|
sleep_intraday_data = pd.read_csv(sensor_data_files["sensor_data"])
|
|
requested_intraday_features = provider["FEATURES"]
|
|
levels_include_all_groups = provider["SLEEP_LEVELS"]["INCLUDE_ALL_GROUPS"]
|
|
requested_sleep_levels = provider["SLEEP_LEVELS"]
|
|
requested_sleep_types = provider["SLEEP_TYPES"]
|
|
|
|
# Name of the features this function can compute
|
|
base_intraday_features = {"LEVELS_AND_TYPES": ["countepisode", "sumduration", "maxduration", "minduration", "avgduration", "medianduration", "stdduration"],
|
|
"RATIOS_TYPE": ["count", "duration"],
|
|
"RATIOS_SCOPE": ["ACROSS_LEVELS", "ACROSS_TYPES", "WITHIN_LEVELS", "WITHIN_TYPES"]}
|
|
base_sleep_levels = {"CLASSIC": ["awake", "restless", "asleep"],
|
|
"STAGES": ["wake", "deep", "light", "rem"],
|
|
"UNIFIED": ["awake", "asleep"]}
|
|
base_sleep_types = ["main", "nap", "all"]
|
|
|
|
# The subset of requested features this function can compute
|
|
intraday_features_to_compute = {key: list(set(requested_intraday_features[key]) & set(base_intraday_features[key])) for key in requested_intraday_features if key in base_intraday_features}
|
|
sleep_levels_to_compute = {key: list(set(requested_sleep_levels[key]) & set(base_sleep_levels[key])) for key in requested_sleep_levels if key in base_sleep_levels}
|
|
sleep_types_to_compute = list(set(requested_sleep_types) & set(base_sleep_types))
|
|
|
|
# Full names
|
|
features_fullnames = featuresFullNames(intraday_features_to_compute, sleep_levels_to_compute, sleep_types_to_compute, levels_include_all_groups)
|
|
sleep_intraday_features = pd.DataFrame(columns=features_fullnames)
|
|
|
|
sleep_intraday_data = filter_data_by_segment(sleep_intraday_data, time_segment)
|
|
|
|
# While level_episode_id is based on levels provided by Fitbit (classic & stages), unified_level_episode_id is based on unified_level.
|
|
sleep_intraday_data.insert(3, "unified_level_episode_id", (sleep_intraday_data[["type_episode_id", "unified_level"]] != sleep_intraday_data[["type_episode_id", "unified_level"]].shift()).any(axis=1).cumsum())
|
|
|
|
if not sleep_intraday_data.empty:
|
|
|
|
sleep_intraday_features = pd.DataFrame()
|
|
|
|
# ALL LEVELS AND TYPES: compute all stats features no matter they are requested or not
|
|
sleep_intraday_features = allStatsFeatures(sleep_intraday_data, base_sleep_levels, base_sleep_types, base_intraday_features["LEVELS_AND_TYPES"], sleep_intraday_features)
|
|
|
|
# RATIOS: only compute requested features
|
|
sleep_intraday_features = ratiosFeatures(sleep_intraday_features, intraday_features_to_compute["RATIOS_TYPE"], intraday_features_to_compute["RATIOS_SCOPE"], sleep_levels_to_compute, sleep_types_to_compute)
|
|
|
|
# Reset index and discard features which are not requested by user
|
|
sleep_intraday_features.index.name = "local_segment"
|
|
sleep_intraday_features.reset_index(inplace=True)
|
|
sleep_intraday_features = sleep_intraday_features[features_fullnames]
|
|
|
|
|
|
return sleep_intraday_features
|