From: Francois Fleuret <francois@fleuret.org>
Date: Tue, 11 May 2021 18:25:15 +0000 (+0200)
Subject: Initial commit.
X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=commitdiff_plain;h=d74d7be5abef26c78d014bd179f2c52f81aca65b;p=pytorch.git

Initial commit.
---

diff --git a/tinymnist.py b/tinymnist.py
new file mode 100755
index 0000000..8642b22
--- /dev/null
+++ b/tinymnist.py
@@ -0,0 +1,93 @@
+#!/usr/bin/env python
+
+# Any copyright is dedicated to the Public Domain.
+# https://creativecommons.org/publicdomain/zero/1.0/
+
+# Written by Francois Fleuret <francois@fleuret.org>
+
+import time, os
+import torch, torchvision
+from torch import nn
+from torch.nn import functional as F
+
+lr, nb_epochs, batch_size = 1e-1, 10, 100
+
+data_dir = os.environ.get('PYTORCH_DATA_DIR') or './data/'
+
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+######################################################################
+
+train_set = torchvision.datasets.MNIST(root = data_dir, train = True, download = True)
+train_input = train_set.data.view(-1, 1, 28, 28).float()
+train_targets = train_set.targets
+
+test_set = torchvision.datasets.MNIST(root = data_dir, train = False, download = True)
+test_input = test_set.data.view(-1, 1, 28, 28).float()
+test_targets = test_set.targets
+
+######################################################################
+
+class SomeLeNet(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.conv1 = nn.Conv2d(1, 32, kernel_size = 5)
+        self.conv2 = nn.Conv2d(32, 64, kernel_size = 5)
+        self.fc1 = nn.Linear(256, 200)
+        self.fc2 = nn.Linear(200, 10)
+
+    def forward(self, x):
+        x = F.relu(F.max_pool2d(self.conv1(x), kernel_size = 3))
+        x = F.relu(F.max_pool2d(self.conv2(x), kernel_size = 2))
+        x = x.view(x.size(0), -1)
+        x = F.relu(self.fc1(x))
+        x = self.fc2(x)
+        return x
+
+######################################################################
+
+model = SomeLeNet()
+
+nb_parameters = sum(p.numel() for p in model.parameters())
+
+print(f'nb_parameters {nb_parameters}')
+
+optimizer = torch.optim.SGD(model.parameters(), lr = lr)
+criterion = nn.CrossEntropyLoss()
+
+model.to(device)
+criterion.to(device)
+
+train_input, train_targets = train_input.to(device), train_targets.to(device)
+test_input, test_targets = test_input.to(device), test_targets.to(device)
+
+mu, std = train_input.mean(), train_input.std()
+train_input.sub_(mu).div_(std)
+test_input.sub_(mu).div_(std)
+
+start_time = time.perf_counter()
+
+for k in range(nb_epochs):
+    acc_loss = 0.
+
+    for input, targets in zip(train_input.split(batch_size),
+                              train_targets.split(batch_size)):
+        output = model(input)
+        loss = criterion(output, targets)
+        acc_loss += loss.item()
+
+        optimizer.zero_grad()
+        loss.backward()
+        optimizer.step()
+
+    nb_test_errors = 0
+    for input, targets in zip(test_input.split(batch_size),
+                              test_targets.split(batch_size)):
+        wta = model(input).argmax(1)
+        nb_test_errors += (wta != targets).long().sum()
+    test_error = nb_test_errors / test_input.size(0)
+    duration = time.perf_counter() - start_time
+
+    print(f'loss {k} {duration:.02f}s {acc_loss:.02f} {test_error*100:.02f}%')
+
+######################################################################