Rename per_mouse to per_session for more clarity

dj_pipeline/notebooks/NP_figures/multi_regression.ipynb

@@ -4,9 +4,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# [Decision Variable Prediction] Figure 5\n",
-    "\n",
-    "## The amount of information available to the mouse correlates inversely with infotaxis behavior."
+    "# The amount of information available to the mouse correlates inversely with infotaxis behavior."
    ]
   },
   {
@@ -67,7 +65,7 @@
     "\n",
     "style()\n",
     "\n",
-    "save_fig_path = \"notebooks/Paper_figures/Figure_output/\""
+    "save_fig_path = \"notebooks/NP_figures/NP_figures/NP_\""
    ]
   },
   {

dj_pipeline/notebooks/Paper_figures/Figure_4.ipynb

@@ -3805,7 +3805,7 @@
     "# interpolated_df_all[\"aperture\"] = interpolated_df_all[\"aperture\"].astype(float)\n",
     "\n",
     "# df_model_all, coef = regression.predict_decision(\n",
-    "#     df=interpolated_df_all, label=model_labels, per_mouse=True\n",
+    "#     df=interpolated_df_all, label=model_labels, per_session=True\n",
     "# )"
    ]
   },
@@ -3845,7 +3845,7 @@
     "    window_pred, window_coef, window_scalers = regression.predict_decision(\n",
     "        df=window_df,\n",
     "        label=model_labels,\n",
-    "        per_mouse=True,\n",
+    "        per_session=True,\n",
     "    )\n",
     "\n",
     "    window_pred[\"model_idx\"] = window_id\n",
@@ -3865,17 +3865,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 89,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "decision_points_all = regression.find_decision_point(df_model_all, \n",
-    "                                                 threshold_uncertainty=0.2)"
+    "# interpolated_df_all[\"aperture\"] = interpolated_df_all[\"aperture\"].astype(float)\n",
+    "\n",
+    "# df_model_all, coef = regression.predict_decision(\n",
+    "#     df=interpolated_df_all, label=model_labels, per_session=True\n",
+    "# )"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 90,
+   "execution_count": null,
    "metadata": {},
    "outputs": [
     {
@@ -3898,48 +3901,46 @@
     }
    ],
    "source": [
-    "fig, ax = plt.subplots(1, 1, figsize=(3, 5))\n",
+    "WINDOW_COUNT = 10\n",
     "\n",
-    "for aperture, color, ypos in zip(decision_points_all.sort_values(\"aperture\").aperture.unique(), \n",
-    "                                     plotting.colors_aperture,\n",
-    "                                     [0.10, 0.05]):\n",
-    "    x = abs(decision_points_all[decision_points_all.aperture==aperture].groupby(\n",
-    "        \"mouse_name\")[\"y\"].mean().values - 27)\n",
-    "    y = trial_df_all[(trial_df_all.aperture==aperture)].groupby(\n",
-    "        \"mouse_name\")[\"trial_rewarded\"].mean().values\n",
-    "\n",
-    "\n",
-    "    # scatter\n",
-    "    ax.scatter(\n",
-    "        x, y,\n",
-    "        color=color,\n",
-    "        alpha=0.7,\n",
-    "        s=50\n",
-    "    )\n",
+    "# Build trial-progress windows like PredictionModel10Windows\n",
+    "df_multi = interpolated_df_all.copy()\n",
+    "df_multi[\"aperture\"] = df_multi[\"aperture\"].astype(float)\n",
+    "df_multi = df_multi.sort_values([\"dataset\", \"trial\", \"trial_length\"]).reset_index(drop=True)\n",
     "\n",
-    "    # regression line\n",
-    "    slope, intercept, r_value, p_value, std_err = stats.linregress(x, y)\n",
-    "    x_fit = np.linspace(x.min(), x.max(), 100)\n",
-    "    y_fit = slope * x_fit + intercept\n",
-    "    ax.plot(x_fit, y_fit, color=color, linestyle=\"--\")\n",
-    "    \n",
-    "    print(f\"Aperture {aperture}: r={r_value}, p={p_value}\")\n",
-    "\n",
-    "    # annotate correlation\n",
-    "    ax.text(\n",
-    "        0.05, ypos,\n",
-    "        f\"r={r_value:.2f}, p={p_value:.3g}\",\n",
-    "        transform=ax.transAxes,\n",
-    "        va=\"top\", ha=\"left\",\n",
-    "        color=color\n",
+    "trial_index = df_multi.groupby([\"dataset\", \"trial\"]).cumcount()\n",
+    "trial_size = df_multi.groupby([\"dataset\", \"trial\"])[\"trial\"].transform(\"size\")\n",
+    "df_multi[\"trial_window\"] = np.clip(\n",
+    "    (trial_index * WINDOW_COUNT / trial_size).astype(int),\n",
+    "    0,\n",
+    "    WINDOW_COUNT - 1,\n",
+    ")\n",
+    "\n",
+    "# Train one model per window (LOGO across datasets)\n",
+    "df_model_parts = []\n",
+    "coef = {}\n",
+    "scalers_by_window = {}\n",
+    "\n",
+    "for window_id in range(WINDOW_COUNT):\n",
+    "    window_df = df_multi[df_multi[\"trial_window\"] == window_id].copy()\n",
+    "    if window_df.empty:\n",
+    "        coef[window_id] = np.array([[np.nan]])\n",
+    "        scalers_by_window[window_id] = []\n",
+    "        continue\n",
+    "\n",
+    "    window_pred, window_coef, window_scalers = regression.predict_decision(\n",
+    "        df=window_df,\n",
+    "        label=model_labels,\n",
+    "        per_session=True,\n",
     "    )\n",
     "\n",
-    "    ax.set_xlabel(\"Distance to screen (cm)\")\n",
-    "    ax.set_ylabel(\"Success rate\")\n",
-    "    ax.set_ylim(0, 1.0)\n",
-    "    sns.despine(offset=10, ax=ax)\n",
+    "    window_pred[\"model_idx\"] = window_id\n",
+    "    df_model_parts.append(window_pred)\n",
+    "\n",
+    "    coef[window_id] = window_coef\n",
+    "    scalers_by_window[window_id] = window_scalers\n",
     "\n",
-    "plt.savefig(save_fig_path + \"figure4_decision_point_vs_performance.svg\", transparent=True)"
+    "df_model_all = pd.concat(df_model_parts, ignore_index=True)"
    ]
   },
   {

dj_pipeline/notebooks/shape-analysis.ipynb

@@ -1106,7 +1106,7 @@
     "df_model, coef, _ = regression.predict_decision(\n",
     "    df=interpolated_j_shaped,\n",
     "    label=model_labels,\n",
-    "    per_mouse=True,\n",
+    "    per_session=True,\n",
     "    scale_data=True,\n",
     "    max_iter=500,\n",
     ")"
@@ -2062,7 +2062,7 @@
     "df_model, coef, _ = regression.predict_decision(\n",
     "    df=interpolated_j_shaped,\n",
     "    label=model_labels,\n",
-    "    per_mouse=True,\n",
+    "    per_session=True,\n",
     "    scale_data=True,\n",
     "    max_iter=500,\n",
     ")"

dj_pipeline/vr4mice/analysis/regression.py

@@ -1,4 +1,5 @@
 import copy
+import warnings
 from typing import List, Optional, Tuple, Union
 
 import matplotlib as mpl
@@ -49,10 +50,11 @@ def predict_decision(
     df,
     label: List[str],
     n_splits: int = 10,
-    per_mouse: bool = True,
+    per_session: bool = True,
     max_iter: int = 100,
     scale_data: bool = True,
     random_state: Optional[int] = None,
+    per_mouse: Optional[bool] = None,
 ) -> Tuple[pd.DataFrame, npt.NDArray, List[Optional[dict]]]:
 
     """Predict the animal's decision based on the `label` data, through a logistic regression.
@@ -61,11 +63,12 @@ def predict_decision(
         df: The dataframe.
         label: A list of column names in the `df` dataframe.
         n_splits: The number of splits fo the cross validation.
-        per_mouse: If `True` split the data per session, else split
-            randomly across all sessions. If per_mouse, we train on
+        per_session: If `True` split the data per session, else split
+            randomly across all sessions. If per_session, we train on
             all sessions but one, and test on the left out session.
         scale_data: If `True`, standardize the data before fitting the model.
         random_state: Random state for reproducibility.
+        per_mouse: Deprecated alias for ``per_session`` kept for backward compatibility.
 
     Returns:
         A tuple of the input dataframe with added ``accuracy`` and ``proba_left`` columns,
@@ -86,6 +89,15 @@ def predict_decision(
         ```
     """
 
+    if per_mouse is not None:
+        warnings.warn(
+            "'per_mouse' is deprecated and will be removed in a future release; "
+            "use 'per_session' instead.",
+            DeprecationWarning,
+            stacklevel=2,
+        )
+        per_session = per_mouse
+
     data = np.asarray(df[label].values)
     labels = df.trial_left_choice.values
 
@@ -98,7 +110,7 @@ def predict_decision(
         max_iter=max_iter, random_state=random_state
     )
 
-    if per_mouse:
+    if per_session:
         sessions = df.dataset.values
         coefs = np.empty((len(np.unique(sessions)), n_features + 1))
         scalers = []

dj_pipeline/vr4mice/schema/decision.py

@@ -423,7 +423,7 @@ class ModelParams(dj.Lookup):
 
 @schema
 class PredictionModel(dj.Computed):
-    """Train logistic regression model per mouse using LOGO cross-validation."""
+    """Train logistic regression model with leave-one-session-out cross-validation."""
 
     definition = """
     -> LabelSet
@@ -432,7 +432,7 @@ class PredictionModel(dj.Computed):
     -> ExperimentStage
     -> vr4mice.Batch
     ---
-    coefficients : <blob>     # coefficients per session (per_mouse=True)
+    coefficients : <blob>     # coefficients per session (per_session=True)
     n_sessions : int32          # number of sessions included
     sessions : <blob>         # list of session dataset names
     random_state : int32        # random state used for reproducibility
@@ -553,7 +553,7 @@ def make(self, key):
             df_model, coef, _ = regression.predict_decision(
                 df=interpolated_df,
                 label=label_set,
-                per_mouse=True,
+                per_session=True,
                 max_iter=params["max_iter"],
                 scale_data=params["scale_data"],
                 random_state=random_state,
@@ -873,7 +873,7 @@ def make(self, key):
                 df_model, coef, scalers = regression.predict_decision(
                     df=window_df,
                     label=label_set,
-                    per_mouse=True,
+                    per_session=True,
                     max_iter=params["max_iter"],
                     scale_data=params["scale_data"],
                     random_state=random_state,