Fill in 0.5 for undefined ratio.

When there are no calls and no smses (of a particular type), the ratio is undefined. But since their number is the same, I argue that the ratio can represent that with a 0.5, similarly to the case where no_calls_all = no_sms_all != 0.

config/environment.yml

@@ -16,6 +16,7 @@ dependencies:
   - python-dotenv
   - pytz
   - seaborn
+  - scikit-learn
   - sqlalchemy
   - statsmodels
   - tabulate

data/example_proximity.csv

@@ -0,0 +1,63 @@
+id,timestamp,device_id,_id,double_proximity,accuracy,label,dateTime
+39017,1565802024310,f67354f7-d675-4b76-80c8-123cc4744a5b,2962,0,3,,2019-08-14T17:00:24Z
+39018,1565802051075,f67354f7-d675-4b76-80c8-123cc4744a5b,2963,0,3,,2019-08-14T17:00:51Z
+39019,1565802051354,f67354f7-d675-4b76-80c8-123cc4744a5b,2964,8,3,,2019-08-14T17:00:51Z
+39089,1565010418305,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,51,5,3,,2019-08-05T13:06:58Z
+39090,1565010772188,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,52,5,3,,2019-08-05T13:12:52Z
+39091,1565012334450,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,53,5,3,,2019-08-05T13:38:54Z
+39092,1565013000660,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,54,5,3,,2019-08-05T13:50:00Z
+39093,1565022742894,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,55,0,3,,2019-08-05T16:32:22Z
+39094,1565089295906,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,56,5,3,,2019-08-06T11:01:35Z
+39095,1565096030817,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,57,0,3,,2019-08-06T12:53:50Z
+39096,1565096367694,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,58,5,3,,2019-08-06T12:59:27Z
+39097,1565096408570,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,59,5,3,,2019-08-06T13:00:08Z
+39098,1565116821528,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,60,5,3,,2019-08-06T18:40:21Z
+39099,1565131345333,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,61,0,3,,2019-08-06T22:42:25Z
+39100,1565131375072,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,62,5,3,,2019-08-06T22:42:55Z
+39101,1565131386353,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,63,0,3,,2019-08-06T22:43:06Z
+39102,1565131389213,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,64,5,3,,2019-08-06T22:43:09Z
+39103,1565131448891,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,65,0,3,,2019-08-06T22:44:08Z
+39104,1565131454131,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,66,5,3,,2019-08-06T22:44:14Z
+39105,1565176143083,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,67,0,3,,2019-08-07T11:09:03Z
+39106,1565179569310,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,68,5,3,,2019-08-07T12:06:09Z
+39107,1565180699173,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,69,5,3,,2019-08-07T12:24:59Z
+39108,1565182538578,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,70,5,3,,2019-08-07T12:55:38Z
+39109,1565192592776,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,71,0,3,,2019-08-07T15:43:12Z
+39110,1565216023797,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,72,5,3,,2019-08-07T22:13:43Z
+39111,1565248358647,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,73,0,3,,2019-08-08T07:12:38Z
+39112,1565275859157,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,74,5,3,,2019-08-08T14:50:59Z
+39113,1565304201431,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,75,0,3,,2019-08-08T22:43:21Z
+39114,1565304229591,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,76,5,3,,2019-08-08T22:43:49Z
+39115,1565304262050,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,77,0,3,,2019-08-08T22:44:22Z
+39116,1565613142970,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,78,5,3,,2019-08-12T12:32:22Z
+39117,1565618266531,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,79,5,3,,2019-08-12T13:57:46Z
+39118,1565618410488,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,80,5,3,,2019-08-12T14:00:10Z
+39119,1565618704942,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,81,5,3,,2019-08-12T14:05:04Z
+39120,1565619005315,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,82,5,3,,2019-08-12T14:10:05Z
+39121,1565619405904,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,83,5,3,,2019-08-12T14:16:45Z
+39122,1565619678037,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,84,5,3,,2019-08-12T14:21:18Z
+39123,1565621206713,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,85,5,3,,2019-08-12T14:46:46Z
+39124,1565626622125,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,86,5,3,,2019-08-12T16:17:02Z
+39125,1565684876738,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,87,5,3,,2019-08-13T08:27:56Z
+39126,1565684956618,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,88,5,3,,2019-08-13T08:29:16Z
+39127,1565684965647,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,89,5,3,,2019-08-13T08:29:25Z
+39128,1565685092246,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,90,5,3,,2019-08-13T08:31:32Z
+39129,1565685136337,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,91,5,3,,2019-08-13T08:32:16Z
+39130,1565685147453,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,92,5,3,,2019-08-13T08:32:27Z
+39131,1565685212523,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,93,5,3,,2019-08-13T08:33:32Z
+39132,1565703397796,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,94,0,3,,2019-08-13T13:36:37Z
+39133,1565776203019,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,95,5,3,,2019-08-14T09:50:03Z
+39134,1565776434168,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,96,5,3,,2019-08-14T09:53:54Z
+39135,1565776435231,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,97,0,3,,2019-08-14T09:53:55Z
+39136,1565776443368,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,98,5,3,,2019-08-14T09:54:03Z
+39137,1565779277109,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,99,0,3,,2019-08-14T10:41:17Z
+39138,1565780016327,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,100,5,3,,2019-08-14T10:53:36Z
+39139,1565780027437,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,101,5,3,,2019-08-14T10:53:47Z
+39140,1565783470934,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,102,5,3,,2019-08-14T11:51:10Z
+39141,1565783801540,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,103,0,3,,2019-08-14T11:56:41Z
+39142,1565783802120,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,104,5,3,,2019-08-14T11:56:42Z
+39143,1565783861495,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,105,5,3,,2019-08-14T11:57:41Z
+39144,1565785318762,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,106,0,3,,2019-08-14T12:21:58Z
+39145,1565785319346,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,107,5,3,,2019-08-14T12:21:59Z
+39146,1565960121019,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,108,5,3,,2019-08-16T12:55:21Z
+39147,1565960226792,fdb06d4a-ee6e-4336-9a96-fc8d2715f243,109,5,3,,2019-08-16T12:57:06Z

exploration/ex_ml_pipeline.py

@@ -0,0 +1,175 @@
+# ---
+# jupyter:
+#   jupytext:
+#     formats: ipynb,py:percent
+#     text_representation:
+#       extension: .py
+#       format_name: percent
+#       format_version: '1.3'
+#       jupytext_version: 1.11.4
+#   kernelspec:
+#     display_name: straw2analysis
+#     language: python
+#     name: straw2analysis
+# ---
+
+# %%
+# %matplotlib inline
+import datetime
+import os
+import sys
+
+import seaborn as sns
+from sklearn import linear_model
+from sklearn.model_selection import LeaveOneGroupOut, cross_val_score
+
+nb_dir = os.path.split(os.getcwd())[0]
+if nb_dir not in sys.path:
+    sys.path.append(nb_dir)
+
+# %%
+import participants.query_db
+from features import esm, helper, proximity
+
+# %% [markdown]
+# # 1. Get the relevant data
+
+# %%
+participants_inactive_usernames = participants.query_db.get_usernames(
+    collection_start=datetime.date.fromisoformat("2020-08-01")
+)
+# Consider only two participants to simplify.
+ptcp_2 = participants_inactive_usernames[0:2]
+
+# %% [markdown]
+# ## 1.1 Labels
+
+# %%
+df_esm = esm.get_esm_data(ptcp_2)
+df_esm_preprocessed = esm.preprocess_esm(df_esm)
+
+# %%
+df_esm_PANAS = df_esm_preprocessed[
+    (df_esm_preprocessed["questionnaire_id"] == 8)
+    | (df_esm_preprocessed["questionnaire_id"] == 9)
+]
+df_esm_PANAS_clean = esm.clean_up_esm(df_esm_PANAS)
+
+# %% [markdown]
+# ## 1.2 Sensor data
+
+# %%
+df_proximity = proximity.get_proximity_data(ptcp_2)
+df_proximity = helper.get_date_from_timestamp(df_proximity)
+df_proximity = proximity.recode_proximity(df_proximity)
+
+# %% [markdown]
+# ## 1.3 Standardization/personalization
+
+# %% [markdown]
+# # 2. Grouping/segmentation
+
+# %%
+df_esm_PANAS_daily_means = (
+    df_esm_PANAS_clean.groupby(["participant_id", "date_lj", "questionnaire_id"])
+    .esm_user_answer_numeric.agg("mean")
+    .reset_index()
+    .rename(columns={"esm_user_answer_numeric": "esm_numeric_mean"})
+)
+
+# %%
+df_esm_PANAS_daily_means = (
+    df_esm_PANAS_daily_means.pivot(
+        index=["participant_id", "date_lj"],
+        columns="questionnaire_id",
+        values="esm_numeric_mean",
+    )
+    .reset_index(col_level=1)
+    .rename(columns={8.0: "PA", 9.0: "NA"})
+    .set_index(["participant_id", "date_lj"])
+)
+
+
+# %%
+df_proximity_daily_counts = proximity.count_proximity(
+    df_proximity, ["participant_id", "date_lj"]
+)
+
+# %%
+df_proximity_daily_counts
+
+# %% [markdown]
+# # 3. Join features (and export to csv?)
+
+# %%
+df_full_data_daily_means = df_esm_PANAS_daily_means.join(
+    df_proximity_daily_counts
+).reset_index()
+
+# %% [markdown]
+# # 4. Machine learning model and parameters
+
+# %%
+lin_reg_proximity = linear_model.LinearRegression()
+
+# %% [markdown]
+# ## 4.1 Validation method
+
+# %%
+logo = LeaveOneGroupOut()
+logo.get_n_splits(
+    df_full_data_daily_means[["freq_prox_near", "prop_prox_near"]],
+    df_full_data_daily_means["PA"],
+    groups=df_full_data_daily_means["participant_id"],
+)
+
+# %% [markdown]
+# ## 4.2 Fit results (export?)
+
+# %%
+cross_val_score(
+    lin_reg_proximity,
+    df_full_data_daily_means[["freq_prox_near", "prop_prox_near"]],
+    df_full_data_daily_means["PA"],
+    groups=df_full_data_daily_means["participant_id"],
+    cv=logo,
+    n_jobs=-1,
+    scoring="r2",
+)
+
+# %%
+lin_reg_proximity.fit(
+    df_full_data_daily_means[["freq_prox_near", "prop_prox_near"]],
+    df_full_data_daily_means["PA"],
+)
+
+# %%
+lin_reg_proximity.score(
+    df_full_data_daily_means[["freq_prox_near", "prop_prox_near"]],
+    df_full_data_daily_means["PA"],
+)
+
+# %% [markdown]
+# # Merging these into a pipeline
+
+# %%
+from machine_learning import pipeline
+
+# %%
+ml_pipeline = pipeline.MachineLearningPipeline(
+    labels_questionnaire="PANAS", data_types="proximity"
+)
+
+# %%
+ml_pipeline.get_labels()
+
+# %% tags=[]
+ml_pipeline.get_sensor_data()
+
+# %%
+ml_pipeline.aggregate_daily()
+
+# %%
+ml_pipeline.df_full_data_daily_means
+
+# %%

exploration/expl_app_categories.py

@@ -0,0 +1,76 @@
+# ---
+# jupyter:
+#   jupytext:
+#     formats: ipynb,py:percent
+#     text_representation:
+#       extension: .py
+#       format_name: percent
+#       format_version: '1.3'
+#       jupytext_version: 1.11.4
+#   kernelspec:
+#     display_name: straw2analysis
+#     language: python
+#     name: straw2analysis
+# ---
+
+# %%
+# %matplotlib inline
+import os
+import sys
+
+import matplotlib.pyplot as plt
+import pandas as pd
+import seaborn as sns
+
+nb_dir = os.path.split(os.getcwd())[0]
+if nb_dir not in sys.path:
+    sys.path.append(nb_dir)
+
+# %%
+from config.models import AppCategories, Participant
+from setup import db_engine, session
+
+# %%
+query_app_categories = session.query(AppCategories)
+with db_engine.connect() as connection:
+    df_app_categories = pd.read_sql(query_app_categories.statement, connection)
+
+# %%
+df_app_categories.head()
+
+# %%
+df_app_categories["play_store_genre"].value_counts()
+
+# %%
+df_category_not_found = df_app_categories[
+    df_app_categories["play_store_genre"] == "not_found"
+]
+
+# %%
+df_category_not_found["play_store_response"].value_counts()
+
+# %%
+df_category_not_found["package_name"].value_counts()
+
+# %%
+manufacturers = [
+    "samsung",
+    "oneplus",
+    "huawei",
+    "xiaomi",
+    "lge",
+    "motorola",
+    "miui",
+    "lenovo",
+    "oppo",
+    "mediatek",
+]
+custom_rom = ["coloros", "lineageos", "myos", "cyanogenmod", "foundation.e"]
+other = ["android", "wssyncmldm"]
+rows_os_manufacturer = df_category_not_found["package_name"].str.contains(
+    "|".join(manufacturers + custom_rom + other), case=False
+)
+
+# %%
+with pd.option_context("display.max_rows", None, "display.max_columns", None):
+    display(df_category_not_found.loc[~rows_os_manufacturer])

exploration/expl_communication.py

@@ -13,14 +13,15 @@
 #     name: straw2analysis
 # ---
 
+# %%
+import importlib
+
 # %%
 # %matplotlib inline
 import os
 import sys
 
 import matplotlib.pyplot as plt
-
-# %%
 import seaborn as sns
 
 nb_dir = os.path.split(os.getcwd())[0]
@@ -28,21 +29,29 @@ if nb_dir not in sys.path:
     sys.path.append(nb_dir)
 
 # %%
-from features.communication import *
+from features import communication, helper
+
+# %%
+importlib.reload(communication)
 
 # %% [markdown]
 # # Example of communication data and feature calculation
 
 # %%
-df_calls = get_call_data(["nokia_0000003"])
+df_calls = communication.get_call_data(["nokia_0000003"])
 print(df_calls)
 
 # %%
-count_comms(df_calls)
+df_calls = helper.get_date_from_timestamp(df_calls)
+communication.count_comms(df_calls, ["date_lj"])
 
 # %%
-df_sms = get_sms_data(["nokia_0000003"])
-count_comms(df_sms)
+df_sms = communication.get_sms_data(["nokia_0000003"])
+df_sms = helper.get_date_from_timestamp(df_sms)
+communication.count_comms(df_sms, ["date_lj"])
+
+# %%
+communication.calls_sms_features(df_calls, df_sms, ["date_lj"])
 
 # %% [markdown]
 # # Call data

features/communication.py

@@ -8,6 +8,43 @@ from setup import db_engine, session
 call_types = {1: "incoming", 2: "outgoing", 3: "missed"}
 sms_types = {1: "received", 2: "sent"}
 
+FEATURES_CALLS = (
+    ["no_calls_all"]
+    + ["no_" + call_type for call_type in call_types.values()]
+    + ["duration_total_" + call_types.get(1), "duration_total_" + call_types.get(2)]
+    + ["duration_max_" + call_types.get(1), "duration_max_" + call_types.get(2)]
+    + ["no_" + call_types.get(1) + "_ratio", "no_" + call_types.get(2) + "_ratio"]
+    + ["no_contacts_calls"]
+)
+
+# FEATURES_CALLS =
+# ["no_calls_all",
+# "no_incoming", "no_outgoing", "no_missed",
+# "duration_total_incoming", "duration_total_outgoing",
+# "duration_max_incoming", "duration_max_outgoing",
+# "no_incoming_ratio", "no_outgoing_ratio",
+# "no_contacts"]
+
+FEATURES_SMS = (
+    ["no_sms_all"]
+    + ["no_" + sms_type for sms_type in sms_types.values()]
+    + ["no_" + sms_types.get(1) + "_ratio", "no_" + sms_types.get(2) + "_ratio"]
+    + ["no_contacts_sms"]
+)
+# FEATURES_SMS =
+# ["no_sms_all",
+#  "no_received", "no_sent",
+#  "no_received_ratio", "no_sent_ratio",
+#  "no_contacts"]
+
+FEATURES_CONTACT = [
+    "proportion_calls_all",
+    "proportion_calls_incoming",
+    "proportion_calls_outgoing",
+    "proportion_calls_contacts",
+    "proportion_calls_missed_sms_received",
+]
+
 
 def get_call_data(usernames: Collection) -> pd.DataFrame:
     """
@@ -98,7 +135,7 @@ def enumerate_contacts(comm_df: pd.DataFrame) -> pd.DataFrame:
     return comm_df
 
 
-def count_comms(comm_df: pd.DataFrame) -> pd.DataFrame:
+def count_comms(comm_df: pd.DataFrame, group_by=None) -> pd.DataFrame:
     """
     Calculate frequencies (and duration) of messages (or calls), grouped by their types.
 
@@ -106,6 +143,9 @@ def count_comms(comm_df: pd.DataFrame) -> pd.DataFrame:
     ----------
     comm_df: pd.DataFrame
         A dataframe of calls or SMSes.
+    group_by: list
+        A list of strings, specifying by which parameters to group.
+        By default, the features are calculated per participant, but could be "date_lj" etc.
 
     Returns
     -------
@@ -118,51 +158,42 @@ def count_comms(comm_df: pd.DataFrame) -> pd.DataFrame:
         * the number of messages by type (received, sent), and
         * the number of communication contacts by type. 
     """
+    if group_by is None:
+        group_by = []
     if "call_type" in comm_df:
+        data_type = "calls"
         comm_counts = (
-            comm_df.value_counts(subset=["participant_id", "call_type"])
-            .unstack()
+            comm_df.value_counts(subset=group_by + ["participant_id", "call_type"])
+            .unstack(level="call_type", fill_value=0)
             .rename(columns=call_types)
             .add_prefix("no_")
         )
         # Count calls by type.
-        comm_counts["no_all"] = comm_counts.sum(axis=1)
+        comm_counts["no_calls_all"] = comm_counts.sum(axis=1)
         # Add a total count of calls.
         comm_counts = comm_counts.assign(
-            no_incoming_ratio=lambda x: x.no_incoming / x.no_all,
-            no_outgoing_ratio=lambda x: x.no_outgoing / x.no_all,
+            no_incoming_ratio=lambda x: x.no_incoming / x.no_calls_all,
+            no_outgoing_ratio=lambda x: x.no_outgoing / x.no_calls_all,
         )
         # Ratio of incoming and outgoing calls to all calls.
         comm_duration_total = (
-            comm_df.groupby(["participant_id", "call_type"])
+            comm_df.groupby(group_by + ["participant_id", "call_type"])
             .sum()["call_duration"]
-            .unstack()
+            .unstack(level="call_type", fill_value=0)
             .rename(columns=call_types)
             .add_prefix("duration_total_")
         )
         # Total call duration by type.
         comm_duration_max = (
-            comm_df.groupby(["participant_id", "call_type"])
+            comm_df.groupby(group_by + ["participant_id", "call_type"])
             .max()["call_duration"]
-            .unstack()
+            .unstack(level="call_type", fill_value=0)
             .rename(columns=call_types)
             .add_prefix("duration_max_")
         )
         # Max call duration by type
-        comm_contacts_counts = (
-            enumerate_contacts(comm_df)
-            .groupby(["participant_id"])
-            .nunique()["contact_id"]
-            .reset_index()
-            .rename(columns={"contact_id": "no_contacts"})
-        )
-        # Number of communication contacts
         comm_features = comm_counts.join(comm_duration_total)
         comm_features = comm_features.join(comm_duration_max)
-        comm_features = comm_features.merge(
-            comm_contacts_counts,
-            on="participant_id"
-            ).set_index("participant_id")
         try:
             comm_features.drop(columns="duration_total_" + call_types[3], inplace=True)
             comm_features.drop(columns="duration_max_" + call_types[3], inplace=True)
@@ -172,33 +203,30 @@ def count_comms(comm_df: pd.DataFrame) -> pd.DataFrame:
         # If there were no missed calls, this exception is raised.
         # But we are dropping the column anyway, so no need to deal with the exception.
     elif "message_type" in comm_df:
+        data_type = "sms"
         comm_counts = (
-            comm_df.value_counts(subset=["participant_id", "message_type"])
-            .unstack()
+            comm_df.value_counts(subset=group_by + ["participant_id", "message_type"])
+            .unstack(level="message_type", fill_value=0)
             .rename(columns=sms_types)
             .add_prefix("no_")
         )
-        comm_counts["no_all"] = comm_counts.sum(axis=1)
+        comm_counts["no_sms_all"] = comm_counts.sum(axis=1)
         # Add a total count of messages.
         comm_features = comm_counts.assign(
-            no_received_ratio=lambda x: x.no_received / x.no_all,
-            no_sent_ratio=lambda x: x.no_sent / x.no_all,
+            no_received_ratio=lambda x: x.no_received / x.no_sms_all,
+            no_sent_ratio=lambda x: x.no_sent / x.no_sms_all,
         )
         # Ratio of incoming and outgoing messages to all messages.
-        comm_contacts_counts = (
-            enumerate_contacts(comm_df)
-            .groupby(["participant_id"])
-            .nunique()["contact_id"]
-            .reset_index()
-            .rename(columns={"contact_id": "no_contacts"})
-        )
-        # Number of communication contacts
-        comm_features = comm_features.merge(
-            comm_contacts_counts,
-            on="participant_id"
-            ).set_index("participant_id")
     else:
         raise KeyError("The dataframe contains neither call_type or message_type")
+    comm_contacts_counts = (
+        enumerate_contacts(comm_df)
+        .groupby(group_by + ["participant_id"])
+        .nunique()["contact_id"]
+        .rename("no_contacts_" + data_type)
+    )
+    # Number of communication contacts
+    comm_features = comm_features.join(comm_contacts_counts)
     return comm_features
 
 
@@ -211,7 +239,7 @@ def contact_features(comm_df: pd.DataFrame) -> pd.DataFrame:
 
     Parameters
     ----------
-    df_enumerated: pd.DataFrame
+    comm_df: pd.DataFrame
         A dataframe of calls or SMSes.
 
     Returns
@@ -221,33 +249,33 @@ def contact_features(comm_df: pd.DataFrame) -> pd.DataFrame:
     """
     df_enumerated = enumerate_contacts(comm_df)
     contacts_count = (
-        df_enumerated
-        .groupby(["participant_id","contact_id"])
-        .size()
-        .reset_index()
+        df_enumerated.groupby(["participant_id", "contact_id"]).size().reset_index()
     )
     # Check whether df contains calls or SMS data since some
-    # features we want to calculate are type-specyfic
+    # features we want to calculate are type-specific
     if "call_duration" in df_enumerated:
         # Add a column with the total duration of calls between two people
         duration_count = (
-            df_enumerated
-            .groupby(["participant_id", "contact_id"])
+            df_enumerated.groupby(["participant_id", "contact_id"])
             # For each participant and for each caller, sum durations of their calls
             ["call_duration"]
             .sum()
             .reset_index()  # Make index (which is actually the participant id) a normal column
             .rename(columns={"call_duration": "total_call_duration"})
         )
-        contacts_count = contacts_count.merge(duration_count, on=["participant_id", "contact_id"])
-        contacts_count.rename(columns={0:"no_calls"}, inplace=True)
+        contacts_count = contacts_count.merge(
+            duration_count, on=["participant_id", "contact_id"]
+        )
+        contacts_count.rename(columns={0: "no_calls"}, inplace=True)
     else:
-        contacts_count.rename(columns={0:"no_sms"}, inplace=True)
+        contacts_count.rename(columns={0: "no_sms"}, inplace=True)
     # TODO:Determine work vs non-work contacts by work hours heuristics
     return contacts_count
 
 
-def calls_sms_features(df_calls: pd.DataFrame, df_sms: pd.DataFrame) -> pd.DataFrame:
+def calls_sms_features(
+    df_calls: pd.DataFrame, df_sms: pd.DataFrame, group_by=None
+) -> pd.DataFrame:
     """
     Calculates additional features relating calls and sms data.
 
@@ -257,13 +285,16 @@ def calls_sms_features(df_calls: pd.DataFrame, df_sms: pd.DataFrame) -> pd.DataF
         A dataframe of calls (return of get_call_data).
     df_sms: pd.DataFrame
         A dataframe of SMSes (return of get_sms_data).
+    group_by: list
+        A list of strings, specifying by which parameters to group.
+        By default, the features are calculated per participant, but could be "date_lj" etc.
 
     Returns
     -------
     df_calls_sms: pd.DataFrame
         The list of features relating calls and sms data for every participant.
         These are:
-        * proportion_calls:
+        * proportion_calls_all:
             proportion of calls in total number of communications
         * proportion_calls_incoming:
             proportion of incoming calls in total number of incoming/received communications
@@ -274,16 +305,22 @@ def calls_sms_features(df_calls: pd.DataFrame, df_sms: pd.DataFrame) -> pd.DataF
         * proportion_calls_contacts:
             proportion of calls contacts in total number of communication contacts
     """
-    count_calls = count_comms(df_calls)
-    count_sms = count_comms(df_sms)
+    if group_by is None:
+        group_by = []
+    count_calls = count_comms(df_calls, group_by)
+    count_sms = count_comms(df_sms, group_by)
     count_joined = (
         count_calls.merge(
-            count_sms, on="participant_id", suffixes=("_calls", "_sms")
-        )  # Merge calls and sms features
-        .reset_index()  # Make participant_id a regular column
+            count_sms,
+            how="outer",
+            left_index=True,
+            right_index=True,
+            validate="one_to_one",
+        )
+        .fillna(0, downcast="infer")
         .assign(
-            proportion_calls=(
-                lambda x: x.no_all_calls / (x.no_all_calls + x.no_all_sms)
+            proportion_calls_all=(
+                lambda x: x.no_calls_all / (x.no_calls_all + x.no_sms_all)
             ),
             proportion_calls_incoming=(
                 lambda x: x.no_incoming / (x.no_incoming + x.no_received)
@@ -295,18 +332,11 @@ def calls_sms_features(df_calls: pd.DataFrame, df_sms: pd.DataFrame) -> pd.DataF
                 lambda x: x.no_outgoing / (x.no_outgoing + x.no_sent)
             ),
             proportion_calls_contacts=(
-                lambda x: x.no_contacts_calls / (x.no_contacts_calls + x.no_contacts_sms)
+                lambda x: x.no_contacts_calls
+                / (x.no_contacts_calls + x.no_contacts_sms)
             )
             # Calculate new features and create additional columns
-        )[
-            [
-                "participant_id",
-                "proportion_calls",
-                "proportion_calls_incoming",
-                "proportion_calls_outgoing",
-                "proportion_calls_contacts",
-                "proportion_calls_missed_sms_received",
-            ]
-        ]  # Filter out only the relevant features
+        )
+        .fillna(0.5, downcast="infer")
     )
     return count_joined

features/esm.py

@@ -1,14 +1,12 @@
-import datetime
 from collections.abc import Collection
 
 import numpy as np
 import pandas as pd
-from pytz import timezone
 
 from config.models import ESM, Participant
+from features import helper
 from setup import db_engine, session
 
-TZ_LJ = timezone("Europe/Ljubljana")
 ESM_STATUS_ANSWERED = 2
 
 GROUP_SESSIONS_BY = ["participant_id", "device_id", "esm_session"]
@@ -67,14 +65,8 @@ def preprocess_esm(df_esm: pd.DataFrame) -> pd.DataFrame:
     df_esm_preprocessed: pd.DataFrame
         A dataframe with added columns: datetime in Ljubljana timezone and all fields from ESM_JSON column.
     """
-    df_esm["datetime_lj"] = df_esm["double_esm_user_answer_timestamp"].apply(
-        lambda x: datetime.datetime.fromtimestamp(x / 1000.0, tz=TZ_LJ)
-    )
-    df_esm = df_esm.assign(
-        date_lj=lambda x: (x.datetime_lj - datetime.timedelta(hours=4)).dt.date
-    )
-    # Since daytime EMAs could *theoretically* last beyond midnight, but never after 4 AM,
-    # the datetime is first translated to 4 h earlier.
+    df_esm = helper.get_date_from_timestamp(df_esm)
+
     df_esm_json = pd.json_normalize(df_esm["esm_json"]).drop(
         columns=["esm_trigger"]
     )  # The esm_trigger column is already present in the main df.
@@ -256,9 +248,9 @@ def clean_up_esm(df_esm_preprocessed: pd.DataFrame) -> pd.DataFrame:
         ESM.ESM_TYPE.get("scale"),
         ESM.ESM_TYPE.get("number"),
     ]
-    df_esm_clean[df_esm_clean["esm_type"].isin(esm_type_numeric)] = df_esm_clean[
+    df_esm_clean.loc[
         df_esm_clean["esm_type"].isin(esm_type_numeric)
-    ].assign(
+    ] = df_esm_clean.loc[df_esm_clean["esm_type"].isin(esm_type_numeric)].assign(
         esm_user_answer_numeric=lambda x: x.esm_user_answer.str.slice(stop=1).astype(
             int
         )

features/helper.py

@@ -0,0 +1,41 @@
+import datetime
+
+import pandas as pd
+from pytz import timezone
+
+TZ_LJ = timezone("Europe/Ljubljana")
+COLUMN_TIMESTAMP = "timestamp"
+COLUMN_TIMESTAMP_ESM = "double_esm_user_answer_timestamp"
+
+
+def get_date_from_timestamp(df_aware) -> pd.DataFrame:
+    """
+    Transform a UNIX timestamp into a datetime (with Ljubljana timezone).
+    Additionally, extract only the date part, where anything until 4 AM is considered the same day.
+
+    Parameters
+    ----------
+    df_aware: pd.DataFrame
+        Any AWARE-type data as defined in models.py.
+
+    Returns
+    -------
+    df_aware: pd.DataFrame
+        The same dataframe with datetime_lj and date_lj columns added.
+
+    """
+    if COLUMN_TIMESTAMP_ESM in df_aware:
+        column_timestamp = COLUMN_TIMESTAMP_ESM
+    else:
+        column_timestamp = COLUMN_TIMESTAMP
+
+    df_aware["datetime_lj"] = df_aware[column_timestamp].apply(
+        lambda x: datetime.datetime.fromtimestamp(x / 1000.0, tz=TZ_LJ)
+    )
+    df_aware = df_aware.assign(
+        date_lj=lambda x: (x.datetime_lj - datetime.timedelta(hours=4)).dt.date
+    )
+    # Since daytime EMAs could *theoretically* last beyond midnight, but never after 4 AM,
+    # the datetime is first translated to 4 h earlier.
+
+    return df_aware

features/proximity.py

@@ -5,6 +5,8 @@ import pandas as pd
 from config.models import Participant, Proximity
 from setup import db_engine, session
 
+FEATURES_PROXIMITY = ["freq_prox_near", "prop_prox_near"]
+
 
 def get_proximity_data(usernames: Collection) -> pd.DataFrame:
     """
@@ -28,3 +30,65 @@ def get_proximity_data(usernames: Collection) -> pd.DataFrame:
     with db_engine.connect() as connection:
         df_proximity = pd.read_sql(query_proximity.statement, connection)
     return df_proximity
+
+
+def recode_proximity(df_proximity: pd.DataFrame) -> pd.DataFrame:
+    """
+    This function recodes proximity from a double to a boolean value.
+    Different proximity sensors report different values,
+        but in our data only several distinct values have ever been found.
+    These are therefore converted into "near" and "far" binary values.
+    See expl_proximity.ipynb for additional info.
+
+    Parameters
+    ----------
+    df_proximity: pd.DataFrame
+        A dataframe of proximity data.
+
+    Returns
+    -------
+    df_proximity: pd.DataFrame
+        The same dataframe with an additional column bool_prox_near,
+            indicating whether "near" proximity was reported.
+        False values correspond to "far" reported by this sensor.
+
+    """
+    df_proximity = df_proximity.assign(bool_prox_near=lambda x: x.double_proximity == 0)
+    return df_proximity
+
+
+def count_proximity(
+    df_proximity: pd.DataFrame, group_by: Collection = None
+) -> pd.DataFrame:
+    """
+    The function counts how many times a "near" value occurs in proximity
+        and calculates the proportion of this counts to all proximity values (i.e. relative count).
+
+    Parameters
+    ----------
+    df_proximity: pd.DataFrame
+        A dataframe of proximity data.
+    group_by: Collection
+        A list of strings, specifying by which parameters to group.
+        By default, the features are calculated per participant, but could be "date_lj" etc.
+
+    Returns
+    -------
+    df_proximity_features: pd.DataFrame
+        A dataframe with the count of "near" proximity values and their relative count.
+    """
+    if group_by is None:
+        group_by = ["participant_id"]
+    if "bool_prox_near" not in df_proximity:
+        df_proximity = recode_proximity(df_proximity)
+    df_proximity["bool_prox_far"] = ~df_proximity["bool_prox_near"]
+    df_proximity_features = df_proximity.groupby(group_by).sum()[
+        ["bool_prox_near", "bool_prox_far"]
+    ]
+    df_proximity_features = df_proximity_features.assign(
+        prop_prox_near=lambda x: x.bool_prox_near / (x.bool_prox_near + x.bool_prox_far)
+    )
+    df_proximity_features = df_proximity_features.rename(
+        columns={"bool_prox_near": "freq_prox_near"}
+    ).drop(columns="bool_prox_far", inplace=False)
+    return df_proximity_features

machine_learning/__init__.py

@@ -0,0 +1,7 @@
+QUESTIONNAIRE_IDS = {"PANAS": {"PA": 8.0, "NA": 9.0}}
+
+QUESTIONNAIRE_IDS_RENAME = {}
+
+for questionnaire in QUESTIONNAIRE_IDS.items():
+    for k, v in questionnaire[1].items():
+        QUESTIONNAIRE_IDS_RENAME[v] = k

machine_learning/pipeline.py

@@ -0,0 +1,125 @@
+import datetime
+
+import pandas as pd
+from sklearn.model_selection import cross_val_score
+
+import participants.query_db
+from features import esm, helper, proximity
+from machine_learning import QUESTIONNAIRE_IDS, QUESTIONNAIRE_IDS_RENAME
+
+
+class MachineLearningPipeline:
+    def __init__(
+        self,
+        labels_questionnaire,
+        labels_scale,
+        data_types,
+        participants_usernames=None,
+        feature_names=None,
+        grouping_variable=None,
+    ):
+        if participants_usernames is None:
+            participants_usernames = participants.query_db.get_usernames(
+                collection_start=datetime.date.fromisoformat("2020-08-01")
+            )
+        self.participants_usernames = participants_usernames
+        self.labels_questionnaire = labels_questionnaire
+        self.data_types = data_types
+
+        if feature_names is None:
+            self.feature_names = []
+        self.df_features = pd.DataFrame()
+        self.labels_scale = labels_scale
+        self.df_labels = pd.DataFrame()
+        self.grouping_variable = grouping_variable
+        self.df_groups = pd.DataFrame()
+
+        self.model = None
+        self.validation_method = None
+
+        self.df_esm = pd.DataFrame()
+        self.df_esm_preprocessed = pd.DataFrame()
+        self.df_esm_interest = pd.DataFrame()
+        self.df_esm_clean = pd.DataFrame()
+
+        self.df_proximity = pd.DataFrame()
+
+        self.df_full_data_daily_means = pd.DataFrame()
+        self.df_esm_daily_means = pd.DataFrame()
+        self.df_proximity_daily_counts = pd.DataFrame()
+
+    def get_labels(self):
+        self.df_esm = esm.get_esm_data(self.participants_usernames)
+        self.df_esm_preprocessed = esm.preprocess_esm(self.df_esm)
+        if self.labels_questionnaire == "PANAS":
+            self.df_esm_interest = self.df_esm_preprocessed[
+                (
+                    self.df_esm_preprocessed["questionnaire_id"]
+                    == QUESTIONNAIRE_IDS.get("PANAS").get("PA")
+                )
+                | (
+                    self.df_esm_preprocessed["questionnaire_id"]
+                    == QUESTIONNAIRE_IDS.get("PANAS").get("NA")
+                )
+            ]
+        self.df_esm_clean = esm.clean_up_esm(self.df_esm_interest)
+
+    def get_sensor_data(self):
+        if "proximity" in self.data_types:
+            self.df_proximity = proximity.get_proximity_data(
+                self.participants_usernames
+            )
+            self.df_proximity = helper.get_date_from_timestamp(self.df_proximity)
+            self.df_proximity = proximity.recode_proximity(self.df_proximity)
+
+    def aggregate_daily(self):
+        self.df_esm_daily_means = (
+            self.df_esm_clean.groupby(["participant_id", "date_lj", "questionnaire_id"])
+            .esm_user_answer_numeric.agg("mean")
+            .reset_index()
+            .rename(columns={"esm_user_answer_numeric": "esm_numeric_mean"})
+        )
+        self.df_esm_daily_means = (
+            self.df_esm_daily_means.pivot(
+                index=["participant_id", "date_lj"],
+                columns="questionnaire_id",
+                values="esm_numeric_mean",
+            )
+            .reset_index(col_level=1)
+            .rename(columns=QUESTIONNAIRE_IDS_RENAME)
+            .set_index(["participant_id", "date_lj"])
+        )
+        self.df_full_data_daily_means = self.df_esm_daily_means.copy()
+        if "proximity" in self.data_types:
+            self.df_proximity_daily_counts = proximity.count_proximity(
+                self.df_proximity, ["participant_id", "date_lj"]
+            )
+            self.df_full_data_daily_means = self.df_full_data_daily_means.join(
+                self.df_proximity_daily_counts
+            )
+
+    def assign_columns(self):
+        self.df_features = self.df_full_data_daily_means[self.feature_names]
+        self.df_labels = self.df_full_data_daily_means[self.labels_scale]
+        if self.grouping_variable:
+            self.df_groups = self.df_full_data_daily_means[self.grouping_variable]
+        else:
+            self.df_groups = None
+
+    def validate_model(self):
+        if self.model is None:
+            raise AttributeError(
+                "Please, specify a machine learning model first, by setting the .model attribute."
+            )
+        if self.validation_method is None:
+            raise AttributeError(
+                "Please, specify a cross validation method first, by setting the .validation_method attribute."
+            )
+        cross_val_score(
+            estimator=self.model,
+            X=self.df_features,
+            y=self.df_labels,
+            groups=self.df_groups,
+            cv=self.validation_method,
+            n_jobs=-1,
+        )

test/test_communication.py

@@ -5,7 +5,7 @@ import pandas as pd
 from numpy.random import default_rng
 from pandas.testing import assert_series_equal
 
-from features.communication import count_comms, enumerate_contacts, get_call_data
+from features.communication import *
 
 rng = default_rng()
 
@@ -76,10 +76,18 @@ class CallsFeatures(unittest.TestCase):
 
     def test_count_comms_calls(self):
         self.features = count_comms(self.calls)
-        print(self.features)
         self.assertIsInstance(self.features, pd.DataFrame)
+        self.assertCountEqual(self.features.columns.to_list(), FEATURES_CALLS)
 
     def test_count_comms_sms(self):
         self.features = count_comms(self.sms)
-        print(self.features)
         self.assertIsInstance(self.features, pd.DataFrame)
+        self.assertCountEqual(self.features.columns.to_list(), FEATURES_SMS)
+
+    def test_calls_sms_features(self):
+        self.features_call_sms = calls_sms_features(self.calls, self.sms)
+        self.assertIsInstance(self.features_call_sms, pd.DataFrame)
+        self.assertCountEqual(
+            self.features_call_sms.columns.to_list(),
+            FEATURES_CALLS + FEATURES_SMS + FEATURES_CONTACT,
+        )

test/test_proximity.py

@@ -0,0 +1,31 @@
+import unittest
+
+from features.proximity import *
+
+
+class ProximityFeatures(unittest.TestCase):
+    df_proximity = pd.DataFrame()
+    df_proximity_recoded = pd.DataFrame()
+    df_proximity_features = pd.DataFrame()
+
+    @classmethod
+    def setUpClass(cls) -> None:
+        cls.df_proximity = pd.read_csv("../data/example_proximity.csv")
+        cls.df_proximity["participant_id"] = 99
+
+    def test_recode_proximity(self):
+        self.df_proximity_recoded = recode_proximity(self.df_proximity)
+        self.assertIn("bool_prox_near", self.df_proximity_recoded)
+        # Is the recoded column present?
+        self.assertIn(True, self.df_proximity_recoded.bool_prox_near)
+        # Are there "near" values in the data?
+        self.assertIn(False, self.df_proximity_recoded.bool_prox_near)
+        # Are there "far" values in the data?
+
+    def test_count_proximity(self):
+        self.df_proximity_recoded = recode_proximity(self.df_proximity)
+        self.df_proximity_features = count_proximity(self.df_proximity_recoded)
+        print(self.df_proximity_features.columns)
+        self.assertCountEqual(
+            self.df_proximity_features.columns.to_list(), FEATURES_PROXIMITY
+        )