feat: examine LDL sampling (#883)

<!--
Reviews go much faster if the reviewer knows what to focus on! Help them
out, e.g.:
Reviewers can skip X, but should pay attention to Y.
-->

psycop/projects/cvd/cohort_examination/notebooks/ldl_sampling.py

@@ -0,0 +1,134 @@
+import marimo
+
+__generated_with = "0.4.0"
+app = marimo.App(width="full")
+
+
+@app.cell
+def __():
+    import polars as pl
+
+    from psycop.common.feature_generation.loaders.raw.load_lab_results import ldl
+
+    return ldl, pl
+
+
+@app.cell
+def __(pl):
+    from psycop.common.feature_generation.loaders.raw.load_visits import physical_visits
+
+    visits = pl.from_pandas(physical_visits())
+    return physical_visits, visits
+
+
+@app.cell
+def __(ldl, pl):
+    ldl_timestamps = (
+        pl.from_pandas(ldl())
+        .drop("value")
+        .with_columns(pl.col("timestamp").alias("latest_ldl_timestamp"))
+    )
+    ldl_timestamps
+    return (ldl_timestamps,)
+
+
+@app.cell
+def __(ldl_timestamps, visits):
+    import datetime as dt
+
+    from timeseriesflattener import Flattener, PredictionTimeFrame, PredictorSpec, ValueFrame
+    from timeseriesflattener.aggregators import LatestAggregator
+
+    with_latest_ldl = Flattener(
+        predictiontime_frame=PredictionTimeFrame(
+            init_df=visits, entity_id_col_name="dw_ek_borger", timestamp_col_name="timestamp"
+        )
+    ).aggregate_timeseries(
+        specs=[
+            PredictorSpec(
+                value_frame=ValueFrame(
+                    init_df=ldl_timestamps,
+                    entity_id_col_name="dw_ek_borger",
+                    value_timestamp_col_name="timestamp",
+                ),
+                lookbehind_distances=[dt.timedelta(days=365 * 10)],
+                aggregators=[LatestAggregator(timestamp_col_name="timestamp")],
+                fallback=dt.datetime(1979, 1, 1),
+            )
+        ]
+    )
+    return (
+        Flattener,
+        LatestAggregator,
+        PredictionTimeFrame,
+        PredictorSpec,
+        ValueFrame,
+        dt,
+        with_latest_ldl,
+    )
+
+
+@app.cell
+def __(pl, with_latest_ldl):
+    collected_df = with_latest_ldl.df.collect()
+    plot_df = collected_df.with_columns(
+        latest_ldl_age_days=(
+            pl.col("timestamp")
+            - pl.col(
+                "pred_latest_ldl_timestamp_within_0_to_3650_days_latest_fallback_1979-01-01 00:00:00"
+            )
+        ).dt.total_days()
+    )
+    plot_df
+    return collected_df, plot_df
+
+
+@app.cell
+def __(plot_df):
+    import plotnine as pn
+
+    plot = (
+        pn.ggplot(data=plot_df)
+        + pn.stat_ecdf(mapping=pn.aes(x="days_after_ldl"))
+        + pn.scale_color_brewer(type="qual", palette=2)
+        + pn.coord_cartesian(xlim=(0, int(365 * 5)))
+        + pn.ylab("Proportion with at least 1 LDL measurement")
+        + pn.xlab("Days from visit to LDL measurement")
+        + pn.scale_x_continuous(expand=(0, 0))
+        + pn.scale_y_continuous(expand=(0, 0))
+        + pn.theme_bw()
+        + pn.theme(legend_position="bottom")
+        + pn.labs(color="")
+    )
+    return plot, pn
+
+
+@app.cell
+def __(plot_df):
+    n = len(plot_df)
+    return (n,)
+
+
+@app.cell
+def __():
+    import marimo as mo
+
+    max_ldl_age_days = mo.ui.slider(1, 730, value=365, step=1, show_value=True)
+    max_ldl_age_days
+    return max_ldl_age_days, mo
+
+
+@app.cell
+def __(max_ldl_age_days, mo, n, pl, plot_df):
+    n_needing_ldl_measurement = len(
+        plot_df.filter(pl.col("latest_ldl_age_days") > max_ldl_age_days.value)
+    )
+
+    mo.md(
+        f"Out of {n} visits, {n_needing_ldl_measurement} ({round(n_needing_ldl_measurement / n * 100, 0)}%) could use a new LDL measurement"
+    )
+    return (n_needing_ldl_measurement,)
+
+
+if __name__ == "__main__":
+    app.run()

pyproject.toml

@@ -41,6 +41,7 @@ exclude = [
   "**/.*venv*",
   ".tox",
   "build",
+  "**/notebooks/*", # Exclude marimo notebooks
   # don't check files that import synthcity as it is not installed on the main environment
   "psycop/projects/scz_bp/model_training/estimator_steps/synth_data_augmentation.py",
   "psycop/projects/scz_bp/model_training/estimator_steps/test_synth_data_augmentation.py",
@@ -133,6 +134,7 @@ exclude = [
   "venv",
   "__init__.py",
   "docs/conf.py",
+  "**/notebooks/*",
 ]
 # Allow unused variables when underscore-prefixed.
 dummy-variable-rgx = "^(_+|(_+[a-zA-Z0-9_]*[a-zA-Z0-9]+?))$"