ML with RAPIDS and missing values.

exploration/ex_ml_pipeline.py

@@ -27,6 +27,7 @@ import yaml
 from pyprojroot import here
 from sklearn import linear_model
 from sklearn.model_selection import LeaveOneGroupOut, cross_val_score
+from sklearn.impute import SimpleImputer
 
 nb_dir = os.path.split(os.getcwd())[0]
 if nb_dir not in sys.path:
@@ -268,31 +269,112 @@ with open(here("rapids/config.yaml"), "r") as file:
 # %%
 for key in rapids_config.keys():    
     if isinstance(rapids_config[key], dict):  # Remove top-level configs
-        if "PROVIDERS" in rapids_config[key]:  # Retain features (that have providers)
+        if ("PROVIDERS" in rapids_config[key]):  # Retain features (that have providers)
             if rapids_config[key]["PROVIDERS"]:  # Remove non-implemented features
                 for provider in rapids_config[key]["PROVIDERS"]:
-                    if rapids_config[key]["PROVIDERS"][provider][
+                    if rapids_config[key]["PROVIDERS"][provider]["COMPUTE"]:  # Check that the features were actually calculated
-                        "COMPUTE"
-                    ]:  # Check that the features were actually calculated
                         if "FEATURES" in rapids_config[key]["PROVIDERS"][provider]:
                             print(key)
                             print(provider)
                             print(rapids_config[key]["PROVIDERS"][provider]["FEATURES"])
 
 # %%
-features_rapids = pd.read_csv(
+features_rapids = pd.read_csv(here("rapids/data/processed/features/all_participants/all_sensor_features.csv"), parse_dates=["local_segment_start_datetime", "local_segment_end_datetime"])
-    here("rapids/data/processed/features/all_participants/all_sensor_features.csv"),
-    parse_dates=["local_segment_start_datetime", "local_segment_end_datetime"],
-)
 
 # %%
 features_rapids.columns
 
 # %%
-features_rapids = features_rapids.assign(
+features_rapids = features_rapids.assign(date_lj=lambda x: x.local_segment_start_datetime.dt.date)
-    date_lj=lambda x: x.local_segment_start_datetime.dt.date
-)
 
 # %%
 features_rapids["participant_id"] = features_rapids["pid"].str.extract("(\d+)")
 features_rapids["participant_id"] = pd.to_numeric(features_rapids["participant_id"])
+features_rapids.set_index(["participant_id", "date_lj"], inplace=True)
+
+# %%
+with open("../machine_learning/config/minimal_labels.yaml", "r") as file:
+    labels_params = yaml.safe_load(file)
+
+# %%
+labels = machine_learning.labels.Labels(**labels_params)
+labels.set_participants_label("all")
+
+# %%
+labels.aggregate_labels(cached=True)
+labels_read = labels.get_aggregated_labels()
+labels_read = labels_read.reset_index()
+labels_read["date_lj"] = labels_read["date_lj"].dt.date
+labels_read.set_index(["participant_id", "date_lj"], inplace=True)
+# date_lj column is parsed as a date and represented as Timestamp, when read from csv.
+# When calculated, it is represented as date.
+
+# %%
+features_rapids.shape
+
+# %%
+labels_read.shape
+
+# %%
+features_labels = features_rapids.join(labels_read, how="inner").reset_index()
+
+# %%
+features_labels.shape
+
+# %%
+features_labels.columns
+
+# %%
+imputer = SimpleImputer(missing_values=np.nan, strategy='mean')
+
+# %%
+feature_columns = features_labels.columns[6:-3]
+label_column = "NA"
+group_column = "pid"
+
+# %%
+lin_reg_rapids = linear_model.LinearRegression()
+logo = LeaveOneGroupOut()
+logo.get_n_splits(
+    features_labels[feature_columns],
+    features_labels[label_column],
+    groups=features_labels[group_column],
+)
+
+# %%
+cross_val_score(
+    lin_reg_rapids,
+    X=imputer.fit_transform(features_labels[feature_columns]),
+    y=features_labels[label_column],
+    groups=features_labels[group_column],
+    cv=logo,
+    n_jobs=-1,
+    scoring="r2",
+)
+
+# %%
+sns.set(rc={"figure.figsize":(16, 8)})
+sns.heatmap(features_labels[feature_columns].isna(), cbar=False)
+
+# %% [markdown] tags=[]
+# ```yaml
+# ALL_CLEANING_INDIVIDUAL:
+#   PROVIDERS:
+#     RAPIDS:
+#       COMPUTE: True
+#       IMPUTE_SELECTED_EVENT_FEATURES: # Fill NAs with 0 only for event-based features, see table below
+#         COMPUTE: True
+#         MIN_DATA_YIELDED_MINUTES_TO_IMPUTE: 0.33 # Any feature value in a time segment instance with phone data yield > [MIN_DATA_YIELDED_MINUTES_TO_IMPUTE] will be replaced with a zero. 
+#       COLS_NAN_THRESHOLD: 0.3 # Discard columns with missing value ratios higher than [COLS_NAN_THRESHOLD]. Set to 1 to disable
+#       COLS_VAR_THRESHOLD: True # Set to True to discard columns with zero variance
+#       ROWS_NAN_THRESHOLD: 1 # Discard rows with missing value ratios higher than [ROWS_NAN_THRESHOLD]. Set to 1 to disable
+#       DATA_YIELD_FEATURE: RATIO_VALID_YIELDED_HOURS # RATIO_VALID_YIELDED_HOURS or RATIO_VALID_YIELDED_MINUTES
+#       DATA_YIELD_RATIO_THRESHOLD: 0.3 # Discard rows with ratiovalidyieldedhours or ratiovalidyieldedminutes feature less than [DATA_YIELD_RATIO_THRESHOLD]. The feature name is determined by [DATA_YIELD_FEATURE] parameter. Set to 0 to disable
+#       DROP_HIGHLY_CORRELATED_FEATURES:
+#         COMPUTE: False
+#         MIN_OVERLAP_FOR_CORR_THRESHOLD: 0.5
+#         CORR_THRESHOLD: 0.95
+#       SRC_SCRIPT: src/features/all_cleaning_individual/rapids/main.R
+# ```
+
+# %%