Add sleep intraday features with RAPIDS provider

Snakefile

@@ -257,11 +257,16 @@ for provider in config["FITBIT_SLEEP_SUMMARY"]["PROVIDERS"].keys():
         files_to_compute.extend(expand("data/processed/features/{pid}/all_sensor_features.csv", pid=config["PIDS"]))
         files_to_compute.append("data/processed/features/all_participants/all_sensor_features.csv")
 
-# for provider in config["FITBIT_SLEEP_INTRADAY"]["PROVIDERS"].keys():
+for provider in config["FITBIT_SLEEP_INTRADAY"]["PROVIDERS"].keys():
-#     if config["FITBIT_SLEEP_INTRADAY"]["PROVIDERS"][provider]["COMPUTE"]:
+    if config["FITBIT_SLEEP_INTRADAY"]["PROVIDERS"][provider]["COMPUTE"]:
-#         files_to_compute.extend(expand("data/raw/{pid}/fitbit_sleep_intraday_raw.csv", pid=config["PIDS"]))
+        files_to_compute.extend(expand("data/raw/{pid}/fitbit_sleep_intraday_raw.csv", pid=config["PIDS"]))
-#         files_to_compute.extend(expand("data/raw/{pid}/fitbit_sleep_intraday_parsed.csv", pid=config["PIDS"]))
+        files_to_compute.extend(expand("data/raw/{pid}/fitbit_sleep_intraday_parsed.csv", pid=config["PIDS"]))
-#         files_to_compute.extend(expand("data/raw/{pid}/fitbit_sleep_intraday_parsed_with_datetime.csv", pid=config["PIDS"]))
+        files_to_compute.extend(expand("data/interim/{pid}/fitbit_sleep_intraday_episodes_resampled.csv", pid=config["PIDS"]))
+        files_to_compute.extend(expand("data/interim/{pid}/fitbit_sleep_intraday_episodes_resampled_with_datetime.csv", pid=config["PIDS"]))
+        files_to_compute.extend(expand("data/interim/{pid}/fitbit_sleep_intraday_features/fitbit_sleep_intraday_{language}_{provider_key}.csv", pid=config["PIDS"], language=config["FITBIT_SLEEP_INTRADAY"]["PROVIDERS"][provider]["SRC_LANGUAGE"].lower(), provider_key=provider.lower()))
+        files_to_compute.extend(expand("data/processed/features/{pid}/fitbit_sleep_intraday.csv", pid=config["PIDS"]))
+        files_to_compute.extend(expand("data/processed/features/{pid}/all_sensor_features.csv", pid=config["PIDS"]))
+        files_to_compute.append("data/processed/features/all_participants/all_sensor_features.csv")
 
 for provider in config["FITBIT_STEPS_SUMMARY"]["PROVIDERS"].keys():
     if config["FITBIT_STEPS_SUMMARY"]["PROVIDERS"][provider]["COMPUTE"]:

config.yaml

@@ -379,6 +379,32 @@ FITBIT_SLEEP_SUMMARY:
       SRC_FOLDER: "rapids" # inside src/features/fitbit_sleep_summary
       SRC_LANGUAGE: "python"
 
+# See https://www.rapids.science/latest/features/fitbit-sleep-intraday/
+FITBIT_SLEEP_INTRADAY:
+  TABLE: fitbit_data
+  INCLUDE_SLEEP_LATER_THAN: &include_sleep_later_than
+    0 # a number ranged from 0 (midnight) to 1439 (23:59)
+  REFERENCE_TIME: &reference_time
+    MIDNIGHT # chosen from "MIDNIGHT" and "START_OF_THE_SEGMENT"
+  PROVIDERS:
+    RAPIDS:
+      COMPUTE: False
+      FEATURES:    
+        LEVELS_AND_TYPES_COMBINING_ALL: True
+        LEVELS_AND_TYPES: [countepisode, sumduration, maxduration, minduration, avgduration, medianduration, stdduration]
+        RATIOS_TYPE: [count, duration]
+        RATIOS_SCOPE: [ACROSS_LEVELS, ACROSS_TYPES, WITHIN_LEVELS, WITHIN_TYPES]
+        ROUTINE: [starttimefirstmainsleep, endtimelastmainsleep, starttimefirstnap, endtimelastnap]
+      SLEEP_LEVELS: 
+        CLASSIC: [awake, restless, asleep]
+        STAGES: [wake, deep, light, rem]
+        UNIFIED: [awake, asleep]
+      SLEEP_TYPES: [main, nap]
+      INCLUDE_SLEEP_LATER_THAN: *include_sleep_later_than
+      REFERENCE_TIME: *reference_time
+      SRC_FOLDER: "rapids" # inside src/features/fitbit_sleep_intraday
+      SRC_LANGUAGE: "python"
+
 # See https://www.rapids.science/latest/features/fitbit-steps-summary/
 FITBIT_STEPS_SUMMARY:
   TABLE: steps_summary

docs/features/fitbit-sleep-intraday.md

@@ -5,7 +5,7 @@ Sensor parameters description for `[FITBIT_SLEEP_INTRADAY]`:
 |Key                                         | Description |
 |----------------|-----------------------------------------------------------------------------------------------------------------------------------
 |`[TABLE]`| Database table name or file path where the sleep intraday data is stored. The configuration keys in [Device Data Source Configuration](../../setup/configuration/#device-data-source-configuration) control whether this parameter is interpreted as table or file.
-|`[INCLUDE_EPISODES_LATER_THAN]`| All sleep episodes that started after this time will be included in the feature computation. It is a number ranging from 0 (midnight) to 1439 (23:59) which denotes the number of minutes after midnight. If a segment is longer than one day, this value is for every day.
+|`[INCLUDE_SLEEP_LATER_THAN]`| All resampled sleep rows (bin interval: one minute) that started after this time will be included in the feature computation. It is a number ranging from 0 (midnight) to 1439 (23:59) which denotes the number of minutes after midnight. If a segment is longer than one day, this value is for every day.
 |`[REFERENCE_TIME]`| The reference point from which the `[ROUTINE]` features are to be computed. Chosen from `MIDNIGHT` and `START_OF_THE_SEGMENT`, default is `MIDNIGHT`. If you have multiple time segments per day it might be more informative to set this flag to `START_OF_THE_SEGMENT`.
 
 The format of the column(s) containing the Fitbit sensor data can be `JSON` or `PLAIN_TEXT`. The data in `JSON` format is obtained directly from the Fitbit API. We support `PLAIN_TEXT` in case you already parsed your data and don't have access to your participants' Fitbit accounts anymore. If your data is in `JSON` format then summary and intraday data come packed together. 
@@ -23,7 +23,14 @@ We provide examples of the input format that RAPIDS expects, note that both exam
         |a748ee1a-1d0b-4ae9-9074-279a2b6ba524     |{"sleep": [{"awakeCount": 1, "awakeDuration": 3, "awakeningsCount": 8, "dateOfSleep": "2020-10-09", "duration": 19320000, "efficiency": 96, "endTime": "2020-10-09T05:57:30.000", "isMainSleep": true, "logId": 14161136803, "minuteData": [{"dateTime": "00:35:30", "value": "2"}, {"dateTime": "00:36:30", "value": "1"}, {"dateTime": "00:37:30", "value": "1"},...], "minutesAfterWakeup": 0, "minutesAsleep": 309, "minutesAwake": 13, "minutesToFallAsleep": 0, "restlessCount": 7, "restlessDuration": 10, "startTime": "2020-10-09T00:35:30.000", "timeInBed": 322}], "summary": {"totalMinutesAsleep": 309, "totalSleepRecords": 1, "totalTimeInBed": 322}}
     
     === "PLAIN_TEXT"
-        Will update this section later.
+
+        All columns are mandatory, however, all except `local_date_time` and `duration` can be empty if you don't have that data. Just have in mind that some features might be inaccurate or empty as `type_episode_id`, `level`, `is_main_sleep`, and `type` are used for sleep episodes extraction. `type_episode_id` is based on where it is extracted: if it is extracted from the 1st "minutesData" block, the `type_episode_id` field will be 0. Similarly, the kth block will be k-1.
+
+        |type_episode_id  |local_date_time     |duration  |level      |is_main_sleep  |type           |
+        |---------------- |------------------- |--------- |---------- |-------------- |-------------- |
+        |0                |2020-10-07 15:55:00 |60        |awake      |0              |classic        |
+        |0                |2020-10-07 15:56:00 |60        |awake      |0              |classic        |
+        |0                |2020-10-07 15:57:00 |60        |restless   |0              |classic        |
 
 ??? example "Example of the structure of source data with Fitbit’s sleep API Version 1.2"
 
@@ -36,7 +43,14 @@ We provide examples of the input format that RAPIDS expects, note that both exam
         |a748ee1a-1d0b-4ae9-9074-279a2b6ba524     |{"sleep":[{"dateOfSleep":"2020-10-12","duration":28980000,"efficiency":93,"endTime":"2020-10-12T09:34:30.000","infoCode":0,"isMainSleep":true,"levels":{"data":[{"dateTime":"2020-10-12T01:31:00.000","level":"wake","seconds":600},{"dateTime":"2020-10-12T01:41:00.000","level":"light","seconds":60},{"dateTime":"2020-10-12T01:42:00.000","level":"deep","seconds":2340},...], "summary":{"deep":{"count":4,"minutes":63,"thirtyDayAvgMinutes":59},"light":{"count":27,"minutes":257,"thirtyDayAvgMinutes":364},"rem":{"count":5,"minutes":94,"thirtyDayAvgMinutes":58},"wake":{"count":24,"minutes":69,"thirtyDayAvgMinutes":95}}},"logId":26589710673,"minutesAfterWakeup":0,"minutesAsleep":415,"minutesAwake":68,"minutesToFallAsleep":0,"startTime":"2020-10-12T01:31:00.000","timeInBed":483,"type":"stages"}],"summary":{"stages":{"deep":63,"light":257,"rem":94,"wake":69},"totalMinutesAsleep":415,"totalSleepRecords":1,"totalTimeInBed":483}}
     
     === "PLAIN_TEXT"
-        Will update this section later.
+
+        All columns are mandatory, however, all except `local_date_time` and `duration` can be empty if you don't have that data. Just have in mind that some features might be inaccurate or empty as `type_episode_id`, `level`, `is_main_sleep`, and `type` are used for sleep episodes extraction. `type_episode_id` is based on where it is extracted: if it is extracted from the 1st "data" and "shortData" block, the `type_episode_id` field will be 0. Similarly, the kth block will be k-1.
+
+        |type_episode_id  |local_date_time     |duration  |level      |is_main_sleep  |type           |
+        |---------------- |------------------- |--------- |---------- |-------------- |-------------- |
+        |0                |2020-10-10 15:36:30 |60        |restless   |1              |stages         |
+        |0                |2020-10-10 15:37:30 |660       |asleep     |1              |stages         |
+        |0                |2020-10-10 15:48:30 |60        |restless   |1              |stages         |
 
 ## RAPIDS provider
 
@@ -45,10 +59,10 @@ We provide examples of the input format that RAPIDS expects, note that both exam
 
 !!! info "File Sequence"
     ```bash
-    # [might update this section later]
     - data/raw/{pid}/fitbit_sleep_intraday_raw.csv
     - data/raw/{pid}/fitbit_sleep_intraday_parsed.csv
-    - data/raw/{pid}/fitbit_sleep_intraday_parsed_with_datetime.csv
+    - data/interim/{pid}/fitbit_sleep_intraday_episodes_resampled.csv
+    - data/interim/{pid}/fitbit_sleep_intraday_episodes_resampled_with_datetime.csv
     - data/interim/{pid}/fitbit_sleep_intraday_features/fitbit_sleep_intraday_{language}_{provider_key}.csv
     - data/processed/features/{pid}/fitbit_sleep_intraday.csv
     ```
@@ -67,13 +81,14 @@ Parameters description for `[FITBIT_SLEEP_INTRADAY][PROVIDERS][RAPIDS]`:
 Features description for `[FITBIT_STEPS_INTRADAY][PROVIDERS][RAPIDS][LEVELS_AND_TYPES]`:
 
 |Feature                                                              |Units          |Description                                                  |
-|-------------------------- |-------------- |-------------------------------------------------------------|
+|------------------------------- |-------------- |-------------------------------------------------------------|
-|countepisode`[LEVEL][TYPE]` |episodes       |Number of `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]`and `[TYPE]`can also be `all` when ``LEVELS_AND_TYPES_COMBINING_ALL`` is True, which groups all `CLASSIC`, `STAGE`, `UNIFIED` levels and both sleep types.
+|countepisode`[LEVEL][TYPE]`     |episodes       |Number of `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]`and `[TYPE]` can also be `all` when ``LEVELS_AND_TYPES_COMBINING_ALL`` is True, which ignores the levels and groups by sleep types.
-|sumduration`[LEVEL][TYPE]`  |minutes        |Total duration of all `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]`and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which groups all `CLASSIC`, `STAGE`, `UNIFIED` levels and both sleep types.
+|sumduration`[LEVEL][TYPE]`      |minutes        |Total duration of all `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]` and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which ignores the levels and groups by sleep types.
-|maxduration`[LEVEL][TYPE]`  |minutes        | Longest duration of any `[LEVEL][TYPE]`sleep episode. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]`and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which groups all `CLASSIC`, `STAGE`, `UNIFIED` levels and both sleep types.
+|maxduration`[LEVEL][TYPE]`      |minutes        | Longest duration of any `[LEVEL][TYPE]`sleep episode. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]` and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which ignores the levels and groups by sleep types.
-|minduration`[LEVEL][TYPE]`  |minutes        | Shortest duration of any `[LEVEL][TYPE]`sleep episode. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]`and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which groups all `CLASSIC`, `STAGE`, `UNIFIED` levels and both sleep types.
+|minduration`[LEVEL][TYPE]`      |minutes        | Shortest duration of any `[LEVEL][TYPE]`sleep episode. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]` and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which ignores the levels and groups by sleep types.
-|avgduration`[LEVEL][TYPE]`  |minutes        | Average duration of all `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]`and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which groups all `CLASSIC`, `STAGE`, `UNIFIED` levels and both sleep types.
+|avgduration`[LEVEL][TYPE]`      |minutes        | Average duration of all `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]` and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which ignores the levels and groups by sleep types.
-|stdduration`[LEVEL][TYPE]`  |minutes        | Standard deviation duration of all `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]`and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which groups all `CLASSIC`, `STAGE`, `UNIFIED` levels and both sleep types.
+|medianduration`[LEVEL][TYPE]`   |minutes        | Median duration of all `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]` and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which ignores the levels and groups by sleep types.
+|stdduration`[LEVEL][TYPE]`      |minutes        | Standard deviation duration of all `[LEVEL][TYPE]`sleep episodes. `[LEVEL]`is one of `[SLEEP_LEVELS]` (e.g. awake-classic or rem-stages) and `[TYPE]` is one of `[SLEEP_TYPES]` (e.g. main). Both `[LEVEL]` and `[TYPE]`can also be `all` when `LEVELS_AND_TYPES_COMBINING_ALL` is True, which ignores the levels and groups by sleep types.
 
 
 Features description for `[FITBIT_STEPS_INTRADAY][PROVIDERS][RAPIDS]` RATIOS `[ACROSS_LEVELS]`:
@@ -122,7 +137,7 @@ Features description for `[FITBIT_STEPS_INTRADAY][PROVIDERS][RAPIDS][ROUTINE]`:
 
 
 !!! note "Assumptions/Observations"
-    1. Deleting values from `[SLEEP_LEVELS]` or `[SLEEP_TYPES]` will only change the features you receive from `[LEVELS_AND_TYPES]`. For example if `STAGES` only contains `[rem, light]` you will not receive `countepisode[wake|deep][TYPE]` or sum, max, min, avg, or std `duration`. These values will not influence `RATIOS` or `ROUTINE` features.
+    1. Deleting values from `[SLEEP_LEVELS]` or `[SLEEP_TYPES]` will only change the features you receive from `[LEVELS_AND_TYPES]`. For example if `STAGES` only contains `[rem, light]` you will not receive `countepisode[wake|deep][TYPE]` or sum, max, min, avg, median, or std `duration`. These values will not influence `RATIOS` or `ROUTINE` features.
     2. Any `[LEVEL]` grouping is done within the elements of each class `CLASSIC`, `STAGES`, and `UNIFIED`. That is, we never combine `CLASSIC` or `STAGES` types to compute features when `LEVELS_AND_TYPES_COMBINING_ALL` is True or when computing `RATIOS`.
     
 

rules/features.smk

@@ -660,6 +660,40 @@ rule fitbit_sleep_summary_r_features:
     script:
         "../src/features/entry.R"
 
+rule resample_sleep_episodes:
+    input:
+        "data/raw/{pid}/fitbit_sleep_intraday_parsed.csv"
+    output:
+        "data/interim/{pid}/fitbit_sleep_intraday_episodes_resampled.csv"
+    script:
+        "../src/features/utils/resample_episodes.R"
+
+rule fitbit_sleep_intraday_python_features:
+    input:
+        sensor_data = "data/interim/{pid}/fitbit_sleep_intraday_episodes_resampled_with_datetime.csv",
+        time_segments_labels = "data/interim/time_segments/{pid}_time_segments_labels.csv"
+    params:
+        provider = lambda wildcards: config["FITBIT_SLEEP_INTRADAY"]["PROVIDERS"][wildcards.provider_key.upper()],
+        provider_key = "{provider_key}",
+        sensor_key = "fitbit_sleep_intraday"
+    output:
+        "data/interim/{pid}/fitbit_sleep_intraday_features/fitbit_sleep_intraday_python_{provider_key}.csv"
+    script:
+        "../src/features/entry.py"
+
+rule fitbit_sleep_intraday_r_features:
+    input:
+        sensor_data = "data/interim/{pid}/fitbit_sleep_intraday_episodes_resampled_with_datetime.csv",
+        time_segments_labels = "data/interim/time_segments/{pid}_time_segments_labels.csv"
+    params:
+        provider = lambda wildcards: config["FITBIT_SLEEP_INTRADAY"]["PROVIDERS"][wildcards.provider_key.upper()],
+        provider_key = "{provider_key}",
+        sensor_key = "fitbit_sleep_intraday"
+    output:
+        "data/interim/{pid}/fitbit_sleep_intraday_features/fitbit_sleep_intraday_r_{provider_key}.csv"
+    script:
+        "../src/features/entry.R"
+
 rule merge_sensor_features_for_individual_participants:
     input:
         feature_files = input_merge_sensor_features_for_individual_participants

src/data/fitbit_parse_sleep.py

@@ -14,66 +14,53 @@ SLEEP_SUMMARY_COLUMNS_V1_2 = ("device_id", "efficiency",
                                 "timestamp")
 SLEEP_SUMMARY_COLUMNS_V1 = SLEEP_SUMMARY_COLUMNS_V1_2 + ("count_awake", "duration_awake", "count_awakenings", "count_restless", "duration_restless")
 
-SLEEP_INTRADAY_COLUMNS = ("device_id",
+SLEEP_INTRADAY_COLUMNS = (# Extract "type_episode_id" field based on summary data: start from 0
+                            "type_episode_id",
+                            "duration",
                             # For "classic" type, original_level is one of {"awake", "restless", "asleep"}
                             # For "stages" type, original_level is one of {"wake", "deep", "light", "rem"}
                             "level",
                             # For "classic" type, unified_level is one of {0, 1} where 0: awake {"awake" + "restless"}, 1: asleep {"asleep"}
                             # For "stages" type, unified_level is one of {0, 1} where 0: awake {"wake"}, 1: asleep {"deep" + "light" + "rem"}
                             "unified_level",
-                            # one of {0, 1} where 0: nap, 1: main sleep
+                            # One of {0, 1} where 0: nap, 1: main sleep
                             "is_main_sleep", 
-                            # one of {"classic", "stages"}
+                            # One of {"classic", "stages"}
                             "type",
                             "local_date_time",
-                            "timestamp")
+                            "start_timestamp",
+                            "end_timestamp")
 
-def mergeLongAndShortData(data_summary):
-    longData = pd.DataFrame(columns=['dateTime', 'level', 'seconds'])
-    shortData = pd.DataFrame(columns=['dateTime','level', 'seconds'])
 
-    windowLength = 30
+def mergeLongAndShortData(data_intraday):
+    long_data = pd.DataFrame(columns=["dateTime", "level"])
+    short_data = pd.DataFrame(columns=["dateTime", "level"])
 
-    for data in data_summary['data']:
+    window_length = 30
-        origEntry = data
+
+    for data in data_intraday["data"]:
         counter = 0
-        numberOfSplits = origEntry['seconds']//windowLength
+        for times in range(data["seconds"] // window_length):
-        for times in range(numberOfSplits):
+            row = {"dateTime": dateutil.parser.parse(data["dateTime"])+timedelta(seconds=counter*window_length), "level": data["level"]}
-            newRow = {'dateTime':dateutil.parser.parse(origEntry['dateTime'])+timedelta(seconds=counter*windowLength),'level':origEntry['level'],'seconds':windowLength}
+            long_data = long_data.append(row, ignore_index = True)
-            longData = longData.append(newRow, ignore_index = True)
             counter = counter + 1
 
-    for data in data_summary['shortData']:
+    for data in data_intraday["shortData"]:
-        origEntry = data
         counter = 0
-        numberOfSplits = origEntry['seconds']//windowLength
+        for times in range(data["seconds"] // window_length):
-        for times in range(numberOfSplits):
+            row = {"dateTime": dateutil.parser.parse(data["dateTime"])+timedelta(seconds=counter*window_length), "level": data["level"]}
-            newRow = {'dateTime':dateutil.parser.parse(origEntry['dateTime'])+timedelta(seconds=counter*windowLength),'level':origEntry['level'],'seconds':windowLength}
+            short_data = short_data.append(row, ignore_index = True)
-            shortData = shortData.append(newRow,ignore_index = True)
             counter = counter + 1
-    longData.set_index('dateTime',inplace=True)
+    long_data.set_index("dateTime",inplace=True)
-    shortData.set_index('dateTime',inplace=True)
+    short_data.set_index("dateTime",inplace=True)
-    longData['level'] = np.where(longData.index.isin(shortData.index) == True,'wake',longData['level'])
+    long_data["level"] = np.where(long_data.index.isin(short_data.index) == True, "wake", long_data["level"])
     
-    longData.reset_index(inplace=True)
+    long_data.reset_index(inplace=True)
     
-    return longData.values.tolist()
+    return long_data.values.tolist()
-
-def classicData1min(data_summary):
-    dataList = list()
-    for data in data_summary['data']:
-        origEntry = data
-        counter = 0
-        timeDuration = 60
-        numberOfSplits = origEntry['seconds']//timeDuration
-        for times in range(numberOfSplits):
-            newRow = {'dateTime':dateutil.parser.parse(origEntry['dateTime'])+timedelta(seconds=counter*timeDuration),'level':origEntry['level'],'seconds':timeDuration}
-            dataList.append(newRow)
-            counter = counter + 1
-    return dataList
 
 # Parse one record for sleep API version 1
-def parseOneRecordForV1(record, device_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type):
+def parseOneRecordForV1(record, device_id, type_episode_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type):
 
     sleep_record_type = "classic"
 
@@ -110,16 +97,16 @@ def parseOneRecordForV1(record, device_id, d_is_main_sleep, records_summary, rec
             d_original_level = SLEEP_CODE2LEVEL[int(data["value"])-1]
 
 
-            row_intraday = (device_id,
+            row_intraday = (type_episode_id, 60,
                             d_original_level, -1, d_is_main_sleep, sleep_record_type,
-                            d_datetime, 0)
+                            d_datetime, 0, 0)
 
             records_intraday.append(row_intraday)
 
     return records_summary, records_intraday
 
 # Parse one record for sleep API version 1.2
-def parseOneRecordForV12(record, device_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type):
+def parseOneRecordForV12(record, device_id, type_episode_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type):
     
     sleep_record_type = record['type']
 
@@ -138,53 +125,25 @@ def parseOneRecordForV12(record, device_id, d_is_main_sleep, records_summary, re
     
     # Intraday data
     if fitbit_data_type == "intraday":
-        if sleep_record_type == 'classic':
+        if sleep_record_type == "classic":
-            start_date = d_start_datetime.date()
+            for data in record["levels"]["data"]:
-            end_date = d_end_datetime.date()
+                d_datetime = dateutil.parser.parse(data["dateTime"])
-            is_before_midnight = True
-            curr_date = start_date
-            data_summary = record['levels']
-            dataSplitted = classicData1min(data_summary)  ##Calling the function to split the data in regular 60 seconds interval
-            for data in dataSplitted:
-                # For overnight episodes, use end_date once we are over midnight
-                d_time = data["dateTime"].time()
-                if is_before_midnight and d_time.hour == 0:
-                    curr_date = end_date
-                d_datetime = datetime.combine(curr_date, d_time)
 
-                d_original_level = data["level"]
+                row_intraday = (type_episode_id, data["seconds"],
-
+                    data["level"], -1, d_is_main_sleep, sleep_record_type,
-                row_intraday = (device_id,
+                    d_datetime, 0, 0)
-                    d_original_level, -1, d_is_main_sleep, sleep_record_type,
-                    d_datetime, 0)
                 records_intraday.append(row_intraday)
         else:
             # For sleep type "stages"
-            start_date = d_start_datetime.date()
+            for data in mergeLongAndShortData(record["levels"]):
-            end_date = d_end_datetime.date()
+                row_intraday = (type_episode_id, 30,
-            is_before_midnight = True
+                    data[1], -1, d_is_main_sleep, sleep_record_type,
-            curr_date = start_date
+                    data[0], 0, 0)
-            data_summary = record['levels']
-            dataList = mergeLongAndShortData(data_summary)
-            for data in dataList:
-
-                d_time = data[0].time()
-                if is_before_midnight and d_time.hour == 0:
-                    curr_date = end_date
-                d_datetime = datetime.combine(curr_date, d_time)
-
-                d_original_level = data[1]
-
-                row_intraday = (device_id,
-                    d_original_level, -1, d_is_main_sleep, sleep_record_type,
-                    d_datetime, 0)
 
                 records_intraday.append(row_intraday)
     
     return records_summary, records_intraday
 
-
-
 def parseSleepData(sleep_data, fitbit_data_type):
     SLEEP_SUMMARY_COLUMNS = SLEEP_SUMMARY_COLUMNS_V1_2
     if sleep_data.empty:
@@ -194,6 +153,7 @@ def parseSleepData(sleep_data, fitbit_data_type):
             return pd.DataFrame(columns=SLEEP_INTRADAY_COLUMNS)
     device_id = sleep_data["device_id"].iloc[0]
     records_summary, records_intraday = [], []
+    type_episode_id = 0
     # Parse JSON into individual records
     for multi_record in sleep_data.fitbit_data:
         for record in json.loads(multi_record)["sleep"]:
@@ -203,11 +163,13 @@ def parseSleepData(sleep_data, fitbit_data_type):
             # For sleep API version 1
             if "awakeCount" in record:
                 SLEEP_SUMMARY_COLUMNS = SLEEP_SUMMARY_COLUMNS_V1
-                records_summary, records_intraday = parseOneRecordForV1(record, device_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type)
+                records_summary, records_intraday = parseOneRecordForV1(record, device_id, type_episode_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type)
             # For sleep API version 1.2
             else:
                 SLEEP_SUMMARY_COLUMNS = SLEEP_SUMMARY_COLUMNS_V1_2
-                records_summary, records_intraday = parseOneRecordForV12(record, device_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type)
+                records_summary, records_intraday = parseOneRecordForV12(record, device_id, type_episode_id, d_is_main_sleep, records_summary, records_intraday, fitbit_data_type)
+            
+            type_episode_id = type_episode_id + 1
 
     if fitbit_data_type == "summary":
         parsed_data = pd.DataFrame(data=records_summary, columns=SLEEP_SUMMARY_COLUMNS)
@@ -216,6 +178,19 @@ def parseSleepData(sleep_data, fitbit_data_type):
 
     return parsed_data
 
+def mergeSleepEpisodes(sleep_data, cols_for_groupby):
+    sleep_episodes = pd.DataFrame(columns=["type_episode_id", "level_episode_id", "level", "unified_level", "is_main_sleep", "type", "start_timestamp", "end_timestamp"])
+    if not sleep_data.empty:
+        sleep_data = sleep_data.groupby(by=cols_for_groupby)
+        sleep_episodes = sleep_data[["start_timestamp"]].first()
+        sleep_episodes["end_timestamp"] = sleep_data["end_timestamp"].last()
+    
+        sleep_episodes.reset_index(inplace=True, drop=False)
+
+    return sleep_episodes
+
+
+
 timezone = snakemake.params["timezone"]
 column_format = snakemake.params["column_format"]
 fitbit_data_type = snakemake.params["fitbit_data_type"]
@@ -237,6 +212,9 @@ elif column_format == "PLAIN_TEXT":
 else:
     raise ValueError("column_format can only be one of ['JSON', 'PLAIN_TEXT'].")
 
+# Drop duplicates
+parsed_data.drop_duplicates(inplace=True)
+
 if parsed_data.shape[0] > 0 and fitbit_data_type == "summary":
     if sleep_episode_timestamp != "start" and sleep_episode_timestamp != "end":
         raise ValueError("SLEEP_EPISODE_TIMESTAMP can only be one of ['start', 'end'].")
@@ -245,6 +223,10 @@ if parsed_data.shape[0] > 0 and fitbit_data_type == "summary":
     
     if not pd.isnull(local_start_date) and not pd.isnull(local_end_date):
         parsed_data = parsed_data.loc[(parsed_data[datetime_column] >= local_start_date) & (parsed_data[datetime_column] < local_end_date)]
+    
+    # Sort by "local_start_date_time" column
+    parsed_data.sort_values(by="local_start_date_time", ascending=True, inplace=True)
+
     parsed_data["timestamp"] = parsed_data[datetime_column].dt.tz_localize(timezone, ambiguous=False, nonexistent="NaT").dropna().astype(np.int64) // 10**6
     parsed_data.dropna(subset=['timestamp'], inplace=True)
     parsed_data.drop(["local_start_date_time", "local_end_date_time"], axis = 1, inplace=True)
@@ -252,8 +234,18 @@ if parsed_data.shape[0] > 0 and fitbit_data_type == "summary":
 if parsed_data.shape[0] > 0 and fitbit_data_type == "intraday":
     if not pd.isnull(local_start_date) and not pd.isnull(local_end_date):
         parsed_data = parsed_data.loc[(parsed_data["local_date_time"] >= local_start_date) & (parsed_data["local_date_time"] < local_end_date)]
-    parsed_data["timestamp"] = parsed_data["local_date_time"].dt.tz_localize(timezone, ambiguous=False, nonexistent="NaT").dropna().astype(np.int64) // 10**6
+    
-    parsed_data.dropna(subset=['timestamp'], inplace=True)
+    # Sort by "local_date_time" column
-    parsed_data["unified_level"] = np.where(parsed_data["level"].isin(["awake", "wake", "restless"]), 0, 1)
+    parsed_data.sort_values(by="local_date_time", ascending=True, inplace=True)
+
+    parsed_data["start_timestamp"] = parsed_data["local_date_time"].dt.tz_localize(timezone, ambiguous=False, nonexistent="NaT").dropna().astype(np.int64) // 10**6
+    parsed_data.dropna(subset=['start_timestamp'], inplace=True)
+    parsed_data["end_timestamp"] = parsed_data["start_timestamp"] + ((parsed_data["duration"] - 1) * 1000) + 999
+    parsed_data["unified_level"] = np.where(parsed_data["level"].isin(["awake", "restless", "wake"]), 0, 1)
+    
+    # Put consecutive rows with the same "level" field together and merge episodes
+    parsed_data.insert(2, "level_episode_id", (parsed_data[["type_episode_id", "level"]] != parsed_data[["type_episode_id", "level"]].shift()).any(axis=1).cumsum())
+    parsed_data = mergeSleepEpisodes(parsed_data, ["type_episode_id", "level_episode_id", "level", "unified_level", "is_main_sleep", "type"])
+
 
 parsed_data.to_csv(snakemake.output[0], index=False)

src/features/fitbit_sleep_intraday/rapids/main.py

@@ -0,0 +1,265 @@
+import pandas as pd
+from datetime import datetime
+import itertools
+
+def featuresFullNames(intraday_features_to_compute, sleep_levels_to_compute, sleep_types_to_compute, consider_all):
+    
+    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 consider_all:
+        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 + ["all"])])
+    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"]:
+        features_fullname.extend(["ratio" + x[0] + x[1] + "within" + x[2] for x in itertools.product(intraday_features_to_compute["RATIOS_TYPE"], sleep_types_to_compute, 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, sleep_types_to_compute)])
+    features_fullname.extend(intraday_features_to_compute["ROUTINE"])
+    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", "start_minutes"])
+
+    if cols_for_groupby and (not sleep_data.empty):
+        sleep_data = sleep_data.groupby(by=cols_for_groupby)
+        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["start_minutes"] = sleep_data["start_minutes"].first()
+    
+        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 + ["all"]):
+        sleep_episodes_classic = sleep_data[sleep_data["is_main_sleep"] == (1 if sleep_type == "main" else 0)] if sleep_type != "all" else sleep_data
+        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 + ["all"]):
+        sleep_episodes_stages = sleep_data[sleep_data["is_main_sleep"] == (1 if sleep_type == "main" else 0)] if sleep_type != "all" else sleep_data
+        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 + ["all"]):
+        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 + ["all"]:
+        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)
+
+    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"
+            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):
+        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:
+            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 singleSleepTypeRoutineFeatures(sleep_intraday_data, routine, reference_time, sleep_type, sleep_intraday_features):
+
+    sleep_intraday_data = sleep_intraday_data[sleep_intraday_data["is_main_sleep"] == (1 if sleep_type == "mainsleep" else 0)]
+    if "starttimefirst" + sleep_type in routine:
+        grouped_first = sleep_intraday_data.groupby(["local_segment"]).first()
+        if reference_time == "MIDNIGHT":
+            sleep_intraday_features["starttimefirst" + sleep_type] = grouped_first["start_minutes"]
+        elif reference_time == "START_OF_THE_SEGMENT":
+            sleep_intraday_features["starttimefirst" + sleep_type] = (grouped_first["start_timestamp"] - grouped_first["segment_start_timestamp"]) / (60 * 1000)
+        else:
+            raise ValueError("Please check FITBIT_SLEEP_INTRADAY section of config.yaml: REFERENCE_TIME can only be MIDNIGHT or START_OF_THE_SEGMENT.")
+    
+    if "endtimelast" + sleep_type in routine:
+        grouped_last = sleep_intraday_data.groupby(["local_segment"]).last()
+        if reference_time == "MIDNIGHT":
+            sleep_intraday_features["endtimelast" + sleep_type] = grouped_last["local_end_date_time"].apply(lambda x: x.hour * 60 + x.minute + x.second / 60)
+        elif reference_time == "START_OF_THE_SEGMENT":
+            sleep_intraday_features["endtimelast" + sleep_type] = (grouped_last["end_timestamp"] - grouped_last["segment_start_timestamp"]) / (60 * 1000)
+        else:
+            raise ValueError("Please check FITBIT_SLEEP_INTRADAY section of config.yaml: REFERENCE_TIME can only be MIDNIGHT or START_OF_THE_SEGMENT.")
+
+    return sleep_intraday_features
+
+def routineFeatures(sleep_intraday_data, routine, reference_time, sleep_type, sleep_intraday_features):        
+    
+    if "starttimefirstmainsleep" in routine or "endtimelastmainsleep" in routine:
+        sleep_intraday_features = singleSleepTypeRoutineFeatures(sleep_intraday_data, routine, reference_time, "mainsleep", sleep_intraday_features)
+    
+    if "starttimefirstnap" in routine or "endtimelastnap" in routine:
+        sleep_intraday_features = singleSleepTypeRoutineFeatures(sleep_intraday_data, routine, reference_time, "nap", sleep_intraday_features)
+    
+    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"])
+
+    consider_all = provider["FEATURES"]["LEVELS_AND_TYPES_COMBINING_ALL"]
+    include_sleep_later_than = provider["INCLUDE_SLEEP_LATER_THAN"]
+    reference_time = provider["REFERENCE_TIME"]
+
+    requested_intraday_features = provider["FEATURES"]
+    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"],
+                                "ROUTINE": ["starttimefirstmainsleep", "endtimelastmainsleep", "starttimefirstnap", "endtimelastnap"]}
+    base_sleep_levels = {"CLASSIC": ["awake", "restless", "asleep"],
+                            "STAGES": ["wake", "deep", "light", "rem"],
+                            "UNIFIED": ["awake", "asleep"]}
+    base_sleep_types = ["main", "nap"]
+
+    # 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, consider_all)
+    sleep_intraday_features = pd.DataFrame(columns=features_fullnames)
+    
+    # Include sleep later than
+    start_minutes = sleep_intraday_data.groupby("start_timestamp").first()["local_time"].apply(lambda x: int(x.split(":")[0]) * 60 + int(x.split(":")[1]) + int(x.split(":")[2]) / 60).to_frame().rename(columns={"local_time": "start_minutes"}).reset_index()
+    sleep_intraday_data = sleep_intraday_data.merge(start_minutes, on="start_timestamp", how="left")
+    sleep_intraday_data = sleep_intraday_data[sleep_intraday_data["start_minutes"] >= include_sleep_later_than]
+
+    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)
+        
+        # ROUTINE: only compute requested features
+        sleep_intraday_features = routineFeatures(sleep_intraday_data, intraday_features_to_compute["ROUTINE"], reference_time, sleep_types_to_compute, sleep_intraday_features)
+
+        # 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

src/features/utils/utils.py

@@ -55,7 +55,7 @@ def chunk_episodes(sensor_episodes):
     sensor_episodes["duration"] = (sensor_episodes["chunked_end_timestamp"] - sensor_episodes["chunked_start_timestamp"]) / (1000 * 60)
 
     # Merge episodes
-    cols_for_groupby = [col for col in sensor_episodes.columns if col not in ["timestamps_segment", "timestamp", "assigned_segments", "start_datetime", "end_datetime", "start_timestamp", "end_timestamp", "duration", "segment_start_timestamp", "segment_end_timestamp", "chunked_start_timestamp", "chunked_end_timestamp"]]
+    cols_for_groupby = [col for col in sensor_episodes.columns if col not in ["timestamps_segment", "timestamp", "assigned_segments", "start_datetime", "end_datetime", "start_timestamp", "end_timestamp", "duration", "chunked_start_timestamp", "chunked_end_timestamp"]]
 
     sensor_episodes_grouped = sensor_episodes.groupby(by=cols_for_groupby)
     merged_sensor_episodes = sensor_episodes_grouped[["duration"]].sum()