🎨 highlight that smoothing slider is both for IQR outlier removal and data smoothing (#54)

* 🎨 highlight that smoothing slider is both for IQR outlier removal and data smoothing

- could be separated
- could be considered to have two sliders

* Split rolling window into separate smoothing and IQR sliders; make smoothing optional

- Replace single `rolling_window` slider with `rolling_window_smoothing` and
  `rolling_window_iqr` so each can be tuned independently
- Add "Apply rolling median smoothing" checkbox; when unchecked the smoothing
  slider is disabled and `df_rolling` passes through filtered data directly
- Update IQR kwargs, session state saves, debug panel, and data dashboard
  label to use the new keys

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01By3g9ScEi2dZh7pYEYCiKV

* Fix black formatting: wrap long f-string lines

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01By3g9ScEi2dZh7pYEYCiKV

* 🔧 update minimum for slider

* Fall back to legacy rolling_window key for backward session compatibility

When restoring a session saved before the slider split, both new sliders
now seed from the old rolling_window value instead of hard-coding 21,
preserving the user's previous preprocessing parameters.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01By3g9ScEi2dZh7pYEYCiKV

---------

Co-authored-by: enryh <noreply@anthropic.com>

Author: enryH

AutoGrowth/0_data_dashboard.py

@@ -22,7 +22,8 @@
 masked = st.session_state.get("masked")
 start_time = st.session_state.get("start_time")
 processing_summary = st.session_state.get("upload_processing_summary_msg")
-rolling_window = st.session_state.get("rolling_window")
+rolling_window_smoothing = st.session_state.get("rolling_window_smoothing")
+apply_smoothing = st.session_state.get("apply_smoothing", True)
 st.session_state.setdefault("yaxis_scale", False)
 st.session_state.setdefault("USE_ELAPSED_TIME_FOR_PLOTS", True)
 use_same_yaxis_scale = bool(st.session_state.get("yaxis_scale", False))
@@ -283,12 +284,12 @@
 if df_rolling is not None:
     with st.container(border=True):
         st.header("Smoothed data view.")
-        if rolling_window is not None:
+        if apply_smoothing and rolling_window_smoothing is not None:
             st.subheader(
-                f"Rolling median in window of {rolling_window}s using filtered OD data"
+                f"Rolling median smoothing (window={rolling_window_smoothing}) applied to filtered OD data"
             )
         else:
-            st.subheader("Filtered raw OD data (untrimmed and not calibrated)")
+            st.subheader("Filtered raw OD data (no smoothing applied)")
         fig = px.scatter(
             df_wide_raw_od_data_filtered,
             x=df_wide_raw_od_data_filtered.index,

AutoGrowth/0_upload_data.py

@@ -413,6 +413,14 @@ def apply_linear_adjustments(
                 ),
             )
         with filter_columns[1]:
+            apply_smoothing = st.checkbox(
+                "Apply rolling median smoothing",
+                value=st.session_state.get("apply_smoothing", True),
+                help=(
+                    "If checked, a rolling median is applied to smooth the OD data "
+                    "before growth rate estimation."
+                ),
+            )
             quantile_max = st.slider(
                 "Max quantile for maximum removal",
                 0.9,
@@ -428,11 +436,21 @@ def apply_linear_adjustments(
                 step=0.1,
                 help="Used when outlier method is IQR. Multiplier of the IQR.",
             )
-            rolling_window = st.slider(
-                "Rolling window (of timepoints) for IQR outlier removal",
+            _rw_fallback = st.session_state.get("rolling_window", 21)
+            rolling_window_smoothing = st.slider(
+                "Rolling window (timepoints) for data smoothing",
+                5,
+                141,
+                st.session_state.get("rolling_window_smoothing", _rw_fallback),
+                step=2,
+                disabled=not apply_smoothing,
+                help="Rolling median window size for OD data smoothing.",
+            )
+            rolling_window_iqr = st.slider(
+                "Rolling window (timepoints) for IQR outlier removal",
                 11,
                 141,
-                st.session_state.get("rolling_window", 21),
+                st.session_state.get("rolling_window_iqr", _rw_fallback),
                 step=2,
                 help="Used when outlier method is IQR.",
             )
@@ -535,7 +553,9 @@ def apply_linear_adjustments(
 st.session_state["outlier_method"] = outlier_method
 st.session_state["quantile_max"] = quantile_max
 st.session_state["iqr_range_value"] = iqr_range_value
-st.session_state["rolling_window"] = rolling_window
+st.session_state["apply_smoothing"] = apply_smoothing
+st.session_state["rolling_window_smoothing"] = rolling_window_smoothing
+st.session_state["rolling_window_iqr"] = rolling_window_iqr
 st.session_state["ecod_factor"] = ecod_factor
 st.session_state["round_time"] = round_time
 st.session_state["aggregate_duplicated_rounded_timepoint"] = (
@@ -717,7 +737,7 @@ def apply_linear_adjustments(
         kwargs_iqr = {
             "method": "iqr",
             "factor": iqr_range_value,
-            "window_size": rolling_window,
+            "window_size": rolling_window_iqr,
         }
         kwargs = (
             kwargs_iqr
@@ -806,15 +826,22 @@ def apply_linear_adjustments(
         # ! should I visualize the values differently?
         df_wide_raw_od_data_filtered = df_wide_raw_od_data_filtered.ffill().bfill()
 
-    df_rolling = (
-        df_wide_raw_od_data_filtered.rolling(
-            rolling_window,
-            min_periods=min_periods,
-            center=True,
+    if apply_smoothing:
+        df_rolling = (
+            df_wide_raw_od_data_filtered.rolling(
+                rolling_window_smoothing,
+                min_periods=min_periods,
+                center=True,
+            )
+            .median()
+            .sort_index()
         )
-        .median()
-        .sort_index()
-    )
+        msg += (
+            f"- Applied rolling median smoothing (window={rolling_window_smoothing}).\n"
+        )
+    else:
+        df_rolling = df_wide_raw_od_data_filtered.sort_index()
+        msg += "- Smoothing disabled; using filtered data directly.\n"
 
     # ? Should it not be possible to be run twice in a single session?
     if od_adjustment_upload is not None:
@@ -861,7 +888,8 @@ def apply_linear_adjustments(
 
     st.session_state["df_rolling"] = df_rolling
 
-    st.session_state["rolling_window"] = int(rolling_window)
+    st.session_state["rolling_window_smoothing"] = int(rolling_window_smoothing)
+    st.session_state["rolling_window_iqr"] = int(rolling_window_iqr)
 
     st.session_state["upload_processing_summary_msg"] = msg
     st.write("### Data processing summary:")
@@ -885,7 +913,11 @@ def apply_linear_adjustments(
                 "filter_by_iqr_range": st.session_state.get("filter_by_iqr_range"),
                 "quantile_max": st.session_state.get("quantile_max"),
                 "iqr_range_value": st.session_state.get("iqr_range_value"),
-                "rolling_window": st.session_state.get("rolling_window"),
+                "apply_smoothing": st.session_state.get("apply_smoothing"),
+                "rolling_window_smoothing": st.session_state.get(
+                    "rolling_window_smoothing"
+                ),
+                "rolling_window_iqr": st.session_state.get("rolling_window_iqr"),
                 "round_time": st.session_state.get("round_time"),
                 "time_ranges": st.session_state.get("time_ranges"),
                 "update_zero_timepoint": st.session_state.get("update_zero_timepoint"),