From 65409455b16d1647c8f2b77600f060286a195442 Mon Sep 17 00:00:00 2001
From: webbrain-one <295484252+webbrain-one@users.noreply.github.com>
Date: Mon, 10 Aug 2026 10:21:56 +0300
Subject: [PATCH] fix: add missing trainer module to resolve import error

Adds trainer.py to fix the ModuleNotFoundError reported in #5.
---
 trainer.py | 55 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 55 insertions(+)
 create mode 100644 trainer.py

diff --git a/trainer.py b/trainer.py
new file mode 100644
index 0000000..d0c6055
--- /dev/null
+++ b/trainer.py
@@ -0,0 +1,55 @@
+import torch
+import torch.nn as nn
+import torch.optim as optim
+from tqdm import tqdm
+import wandb
+
+
+class Trainer:
+    def __init__(self, model, loss_fn, lr=1e-3, T_max=1000, weight_decay=0, wandb_log=False):
+        self.model = model
+        self.loss_fn = loss_fn
+        self.optimizer = optim.AdamW(model.parameters(), lr=lr, weight_decay=weight_decay)
+        self.scheduler = optim.lr_scheduler.CosineAnnealingLR(self.optimizer, T_max=T_max)
+        self.wandb_log = wandb_log
+
+    def plot_architecture(self, dataloader, depth=6):
+        try:
+            from torchview import draw_graph
+            sample_input = next(iter(dataloader))[0]
+            draw_graph(self.model, input_size=(1, *sample_input.shape[1:]), device=next(self.model.parameters()).device).save_graph(fname='model_graph_new.png', max_depth=depth)
+        except Exception as e:
+            print(f"Skipping plot_architecture: {e}")
+
+    def train_loop(self, train_loader, test_loader, n_epochs, val_metric=None):
+        device = next(self.model.parameters()).device
+        self.model.train()
+        for epoch in range(n_epochs):
+            running_loss = 0.0
+            for inputs, labels in tqdm(train_loader, desc=f"Epoch {epoch}"):
+                inputs, labels = inputs.to(device), labels.to(device)
+                self.optimizer.zero_grad()
+                outputs = self.model(inputs)
+                loss = self.loss_fn(outputs, labels)
+                loss.backward()
+                self.optimizer.step()
+                running_loss += loss.item()
+                self.scheduler.step()
+
+            avg_loss = running_loss / len(train_loader)
+            if self.wandb_log:
+                wandb.log({"epoch_loss": avg_loss})
+
+            if test_loader is not None and val_metric is not None:
+                self.model.eval()
+                val_metric.reset()
+                with torch.no_grad():
+                    for inputs, labels in test_loader:
+                        inputs, labels = inputs.to(device), labels.to(device)
+                        outputs = self.model(inputs)
+                        val_metric.update(outputs, labels)
+                val_acc = val_metric.compute().item()
+                print(f"Epoch {epoch}, Loss: {avg_loss:.4f}, Val Acc: {val_acc:.4f}")
+                if self.wandb_log:
+                    wandb.log({"val_acc": val_acc})
+            self.model.train()