|
import torch |
|
import torch.nn as nn |
|
import torch.optim as optim |
|
import math |
|
import numpy as np |
|
from typing import Dict, Optional |
|
from .node import CognitiveNode |
|
|
|
class DynamicCognitiveNet(nn.Module): |
|
"""Arsitektur jaringan dengan manajemen tensor yang robust""" |
|
def __init__(self, input_size: int, output_size: int): |
|
super().__init__() |
|
self.input_size = input_size |
|
self.output_size = output_size |
|
|
|
|
|
self.input_nodes = nn.ModuleList([ |
|
CognitiveNode(i, 1) for i in range(input_size) |
|
]) |
|
self.output_nodes = nn.ModuleList([ |
|
CognitiveNode(input_size + i, 1) for i in range(output_size) |
|
]) |
|
|
|
|
|
self.connections = nn.ParameterDict() |
|
self._init_base_connections() |
|
|
|
|
|
self.emotional_state = nn.Parameter(torch.tensor(0.0)) |
|
self.optimizer = optim.AdamW(self.parameters(), lr=0.001) |
|
self.loss_fn = nn.MSELoss() |
|
|
|
def _init_base_connections(self): |
|
"""Inisialisasi koneksi input-output""" |
|
for in_node in self.input_nodes: |
|
for out_node in self.output_nodes: |
|
conn_id = f"{in_node.id}->{out_node.id}" |
|
self.connections[conn_id] = nn.Parameter( |
|
torch.randn(1) * 0.1 |
|
) |
|
|
|
def forward(self, x: torch.Tensor) -> torch.Tensor: |
|
|
|
x = x.view(-1) |
|
|
|
|
|
activations = {} |
|
for i, node in enumerate(self.input_nodes): |
|
activations[node.id] = node(x[i].unsqueeze(0)) |
|
|
|
|
|
outputs = [] |
|
for out_node in self.output_nodes: |
|
integrated = [] |
|
for in_node in self.input_nodes: |
|
conn_id = f"{in_node.id}->{out_node.id}" |
|
weight = torch.sigmoid(self.connections[conn_id]) |
|
integrated.append(activations[in_node.id] * weight) |
|
|
|
if integrated: |
|
combined = sum(integrated) / math.sqrt(len(integrated)) |
|
outputs.append(out_node(combined)) |
|
|
|
return torch.stack(outputs).squeeze() |
|
|
|
def structural_update(self, global_reward: float): |
|
"""Update struktur jaringan""" |
|
|
|
for conn_id in list(self.connections.keys()): |
|
new_weight = self.connections[conn_id] + 0.1 * global_reward |
|
self.connections[conn_id].data = new_weight.clamp(-1, 1) |
|
|
|
|
|
if global_reward < -0.5: |
|
new_conn = self._find_underutilized_connection() |
|
if new_conn and new_conn not in self.connections: |
|
self.connections[new_conn] = nn.Parameter(torch.randn(1) * 0.1) |
|
|
|
def _find_underutilized_connection(self) -> Optional[str]: |
|
"""Mencari pasangan node yang kurang aktif""" |
|
input_act = {n.id: np.mean(n.recent_activations) |
|
for n in self.input_nodes if n.recent_activations} |
|
output_act = {n.id: np.mean(n.recent_activations) |
|
for n in self.output_nodes if n.recent_activations} |
|
|
|
if not input_act or not output_act: |
|
return None |
|
|
|
src = min(input_act, key=lambda k: input_act[k]) |
|
tgt = min(output_act, key=lambda k: output_act[k]) |
|
return f"{src}->{tgt}" |
|
|
|
def train_step(self, x: torch.Tensor, y: torch.Tensor) -> float: |
|
"""Training step dengan error handling""" |
|
self.optimizer.zero_grad() |
|
|
|
try: |
|
pred = self(x.view(-1)) |
|
loss = self.loss_fn(pred, y.view(-1)) |
|
except Exception as e: |
|
print(f"Error forward: {e}") |
|
return float('nan') |
|
|
|
|
|
reg_loss = sum(p.abs().mean() for p in self.connections.values()) |
|
total_loss = loss + 0.01 * reg_loss |
|
|
|
try: |
|
total_loss.backward() |
|
self.optimizer.step() |
|
except Exception as e: |
|
print(f"Error backward: {e}") |
|
return float('nan') |
|
|
|
|
|
self.emotional_state.data = torch.sigmoid( |
|
self.emotional_state + (0.5 - loss.item()) * 0.1 |
|
) |
|
self.structural_update(0.5 - loss.item()) |
|
|
|
return total_loss.item() |