Upload app_space.py

app_space.py

@@ -2,10 +2,13 @@ import io
 import json
 import os
 import sys
+import math
 from typing import Any, Dict, List
 
 import gradio as gr
 import matplotlib.pyplot as plt
+import numpy as np
+import matplotlib.image as mpimg
 
 # Ensure local package is importable when running in Hugging Face Spaces
 ROOT = os.path.dirname(os.path.abspath(__file__))
@@ -40,8 +43,11 @@ def run_infer(signal_text: str) -> Dict[str, Any]:
     )
     patient_context = {"patient_id": "demo"}
     rules_result = evaluate_ecg_rules(patient_context, model_output)
-    explanations = [*(model_output.get("gating", {}).get("explanations", []) if isinstance(model_output.get("gating"), dict) else []),
-                    *rules_result.get("explanations", [])]
+    explanations = [
+        *(model_output.get("gating", {}).get("explanations", []) if isinstance(model_output.get("gating"), dict) else []),
+        *rules_result.get("explanations", []),
+    ]
+    summary = f"Windows kept: {gating_meta.get('selected_windows',0)}/{gating_meta.get('total_windows',0)} • ratio={gating_meta.get('ratio',1):.2f}"
     return {
         "label": model_output.get("label"),
         "score": round(float(model_output.get("score", 0.0)), 3),
@@ -49,6 +55,7 @@ def run_infer(signal_text: str) -> Dict[str, Any]:
         "alert_level": rules_result.get("alert_level", "none"),
         "gated_ratio": round(model_output.get("gated_ratio", 1.0), 3),
         "gating": gating_meta,
+        "gating_summary": summary,
         "explanations": explanations,
     }
 
@@ -66,15 +73,31 @@ def plot_gating(signal_text: str):
     fig.savefig(buf, format="png", dpi=120)
     plt.close(fig)
     buf.seek(0)
-    import numpy as np
-    import matplotlib.image as mpimg
-    buf.seek(0)
     np_img = mpimg.imread(buf)
-    return np_img
-
-
-demo_normal = [0.05 for _ in range(256)]
-demo_afib = [0.3 for _ in range(256)]
+    windows = meta.get("windows", [])
+    table = [
+        {
+            "start": w.get("start"),
+            "end": w.get("end"),
+            "significance": round(float(w.get("significance", 0.0)), 3),
+            "prob": round(float(w.get("probability", 0.0)), 3),
+            "selected": bool(w.get("selected")),
+        }
+        for w in windows
+    ]
+    summary = f"Windows kept: {meta.get('selected_windows',0)}/{meta.get('total_windows',0)} • ratio={meta.get('ratio',1):.2f}"
+    return np_img, summary, table
+
+
+# Demo signals with more structure so gating can skip/keep meaningfully
+demo_normal = [0.05 * math.sin(2 * math.pi * 2 * (i / 256)) for i in range(256)]
+demo_afib = [
+    0.25 * math.sin(2 * math.pi * 6 * (i / 256))
+    + 0.05 * math.sin(2 * math.pi * 15 * (i / 256))
+    + (0.15 if i % 40 == 0 else 0.0)
+    for i in range(256)
+]
+demo_noise = [0.02 * math.sin(2 * math.pi * 1 * (i / 256)) + (0.01 if i % 13 == 0 else 0.0) for i in range(256)]
 
 with gr.Blocks(title="Sundew ECG Demo") as demo:
     gr.Markdown("### Neurosymbolic ECG • Sundew Gating + Rules")
@@ -93,16 +116,21 @@ with gr.Blocks(title="Sundew ECG Demo") as demo:
                 value=json.dumps(demo_afib[:128]),
             )
             img = gr.Image(type="numpy", label="Raw vs Gated")
+            summary_box = gr.Textbox(label="Gating summary")
+            table = gr.Dataframe(headers=["start", "end", "significance", "prob", "selected"], datatype=["number", "number", "number", "number", "bool"], wrap=True)
             btn2 = gr.Button("Show gating")
-            btn2.click(plot_gating, inputs=inp2, outputs=img)
+            btn2.click(plot_gating, inputs=inp2, outputs=[img, summary_box, table])
         with gr.Tab("Demos"):
             out_demo = gr.JSON()
             btn_n = gr.Button("Normal")
             btn_a = gr.Button("Arrhythmia-ish")
+            btn_noise = gr.Button("Noisy baseline")
             hidden_n = gr.Textbox(value=json.dumps(demo_normal), visible=False)
             hidden_a = gr.Textbox(value=json.dumps(demo_afib), visible=False)
+            hidden_noise = gr.Textbox(value=json.dumps(demo_noise), visible=False)
             btn_n.click(run_infer, inputs=hidden_n, outputs=out_demo)
             btn_a.click(run_infer, inputs=hidden_a, outputs=out_demo)
+            btn_noise.click(run_infer, inputs=hidden_noise, outputs=out_demo)
 
 
 if __name__ == "__main__":