Upload app.py

app.py

@@ -7,8 +7,6 @@ matplotlib.use('Agg')
 import matplotlib.pyplot as plt
 from torchvision import datasets, transforms
 import gradio as gr
-import json
-import os
 import io
 from datetime import datetime
 from PIL import Image
@@ -17,7 +15,6 @@ from PIL import Image
 transform = transforms.Compose([transforms.ToTensor()])
 mnist_dataset = datasets.MNIST(root='./data', train=True, download=False, transform=transform)
 
-HISTORY_FILE = "history.json"
 
 
 class SKAModel(nn.Module):
@@ -128,20 +125,7 @@ def get_mnist_subset(samples_per_class, data_seed=0):
     return torch.stack(images_list)
 
 
-def load_history():
-    if os.path.exists(HISTORY_FILE):
-        with open(HISTORY_FILE, "r") as f:
-            return json.load(f)
-    return []
-
-
-def save_history(history):
-    with open(HISTORY_FILE, "w") as f:
-        json.dump(history, f)
-
-
-def plot_convergence_comparison():
-    history = load_history()
+def plot_convergence_comparison(history):
     if not history:
         fig, ax = plt.subplots()
         ax.text(0.5, 0.5, "No history yet — run at least one architecture.", ha='center', va='center')
@@ -215,7 +199,7 @@ def plot_convergence_comparison():
     return Image.open(buf)
 
 
-def run_ska(n1, n2, n3, n4, K, tau, samples_per_class, data_seed):
+def run_ska(n1, n2, n3, n4, K, tau, samples_per_class, data_seed, history):
     layer_sizes = [int(n1), int(n2), int(n3), int(n4)]
     neurons_str = ", ".join(str(n) for n in layer_sizes)
 
@@ -256,8 +240,6 @@ def run_ska(n1, n2, n3, n4, K, tau, samples_per_class, data_seed):
         for l in range(num_layers)
     ]
 
-    # Save to history
-    history = load_history()
     run = {
         "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
         "architecture": neurons_str,
@@ -268,8 +250,7 @@ def run_ska(n1, n2, n3, n4, K, tau, samples_per_class, data_seed):
         "convergence_state": convergence_state,
         "entropy_history_norm": entropy_history_norm,
     }
-    history.append(run)
-    save_history(history)
+    history = history + [run]
 
     # Plot 1: normalized entropy trajectory (current run)
     fig1, axes1 = plt.subplots(num_layers, 1, figsize=(10, 3 * num_layers), sharex=True)
@@ -284,15 +265,13 @@ def run_ska(n1, n2, n3, n4, K, tau, samples_per_class, data_seed):
     fig1.suptitle(f"Architecture: [{neurons_str}] | K={K} | τ={tau:.2f}", fontsize=12)
     fig1.tight_layout()
 
-    # Plot 2: convergence state comparison across all runs
-    fig2 = plot_convergence_comparison()
+    fig2 = plot_convergence_comparison(history)
 
-    return fig1, fig2
+    return fig1, fig2, history
 
 
 def clear_history():
-    save_history([])
-    return plot_convergence_comparison()
+    return plot_convergence_comparison([])
 
 
 with gr.Blocks(title="SKA Entropy State Explorer") as demo:
@@ -336,10 +315,12 @@ with gr.Blocks(title="SKA Entropy State Explorer") as demo:
             plot_current = gr.Plot(label="Current Run: Normalized Entropy Trajectory")
             plot_comparison = gr.Image(label="4D Entropy State Trajectory")
 
+    history_state = gr.State([])
+
     run_btn.click(
         fn=run_ska,
-        inputs=[n1_input, n2_input, n3_input, n4_input, k_slider, tau_slider, samples_slider, seed_slider],
-        outputs=[plot_current, plot_comparison],
+        inputs=[n1_input, n2_input, n3_input, n4_input, k_slider, tau_slider, samples_slider, seed_slider, history_state],
+        outputs=[plot_current, plot_comparison, history_state],
     )
     clear_btn.click(
         fn=clear_history,