Update evo_transformer.py

evo_transformer.py

@@ -1,10 +1,36 @@
 import random
-import copy
+import json
+import csv
+import io
 
 class EvoTransformer:
     def __init__(self):
         self.history = []
-        self.base_config = {
+        self.current = {}
+
+    def reset(self):
+        self.history = []
+        self.current = {}
+
+    def mutate(self, config):
+        config["layers"] = max(1, config["layers"] + random.choice([-1, 1]))
+        config["attention_heads"] = max(1, config["attention_heads"] + random.choice([-1, 1]))
+        config["ffn_dim"] = max(64, config["ffn_dim"] + random.choice([-256, 256]))
+        config["dropout"] = min(0.5, max(0.01, config["dropout"] + random.choice([-0.02, 0.02])))
+        config["memory"] = random.choice([True, False])
+        return config
+
+    def simulate_accuracy(self, config):
+        return round(random.uniform(70, 90), 2)
+
+    def estimate_parameters(self, config):
+        return round(
+            config["layers"] * config["attention_heads"] * config["ffn_dim"] / 1e6,
+            2
+        )
+
+    def evolve(self, generations):
+        base_config = {
             "layers": 4,
             "attention_heads": 4,
             "ffn_dim": 1024,
@@ -12,26 +38,31 @@ class EvoTransformer:
             "memory": False
         }
 
-    def reset(self):
-        self.history = []
+        for _ in range(generations):
+            base_config = self.mutate(base_config.copy())
+            acc = self.simulate_accuracy(base_config)
+            params = self.estimate_parameters(base_config)
+            record = {"traits": base_config.copy(), "accuracy": acc, "parameters": params}
+            self.history.append(record)
 
-    def mutate(self, config):
-        new_config = copy.deepcopy(config)
-        if random.random() < 0.5:
-            new_config["layers"] = min(12, max(1, new_config["layers"] + random.choice([-1, 1])))
-        if random.random() < 0.5:
-            new_config["attention_heads"] = min(12, max(1, new_config["attention_heads"] + random.choice([-1, 1])))
-        if random.random() < 0.5:
-            new_config["ffn_dim"] = min(4096, max(128, new_config["ffn_dim"] + random.choice([-512, 512])))
-        if random.random() < 0.5:
-            new_config["dropout"] = round(min(0.5, max(0.0, new_config["dropout"] + random.choice([-0.02, 0.02]))), 2)
-        if random.random() < 0.3:
-            new_config["memory"] = not new_config["memory"]
-        return new_config
-
-    def run_evolution(self, generations=5):
-        current = self.base_config
-        self.history.append(current)
-        for _ in range(generations - 1):
-            current = self.mutate(current)
-            self.history.append(current)
+        self.current = self.history[-1]
+
+    def get_best_config(self):
+        return self.current
+
+    def export_csv(self):
+        output = io.StringIO()
+        writer = csv.writer(output)
+        writer.writerow(["Generation", "Layers", "Attention Heads", "FFN Dim", "Dropout", "Memory", "Accuracy", "Params"])
+        for i, entry in enumerate(self.history):
+            t = entry["traits"]
+            writer.writerow([
+                i+1,
+                t["layers"], t["attention_heads"], t["ffn_dim"],
+                t["dropout"], t["memory"],
+                entry["accuracy"], entry["parameters"]
+            ])
+        return output.getvalue()
+
+    def export_json(self):
+        return json.dumps(self.history, indent=2)