pytorch introduction code

Author: pythonlessons

CHANGELOG.md

@@ -1,4 +1,4 @@
-## [1.0.0] - 2022-02-27
+## [1.0.0] - 2022-03-03
 
 ### Changed
 - detaching TensorFlow from mltu, now mltu is only a collection of utilities for training machine learning models

Tutorials/05_sound_to_text/README.md

@@ -0,0 +1,10 @@
+# Introduction to speech recognition with TensorFlow
+## Master the basics of speech recognition with TensorFlow: Learn how to build and train models, implement real-time audio recognition, and develop practical applications
+
+
+## **Detailed tutorial**:
+## [Introduction to speech recognition with TensorFlow](https://pylessons.com/speech-recognition)
+
+<p align="center">
+  <img src="https://pylessons.com/media/Tutorials/TensorFlow-CAPTCHA-solver/speech-recognition/speech-recognition.png">
+</p>

Tutorials/06_pytorch_introduction/README.md

@@ -0,0 +1,9 @@
+# Introduction to speech recognition with TensorFlow
+## Master the basics of speech recognition with TensorFlow: Learn how to build and train models, implement real-time audio recognition, and develop practical applications<br><br>
+
+# **Detailed tutorial**:
+## [Introduction to PyTorch in a practical way](https://pylessons.com/pytorch-introduction)
+
+<p align="center">
+  <img src="https://pylessons.com/media/Tutorials/TensorFlow-CAPTCHA-solver/pytorch-introduction/pytorch-introduction.png">
+</p>

Tutorials/06_pytorch_introduction/model.py

@@ -1,6 +1,7 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
+# Define the model architecture
 class Net(nn.Module):
     def __init__(self):
         super(Net, self).__init__()
@@ -17,4 +18,5 @@ def forward(self, x):
         x = F.relu(self.fc1(x))
         x = F.dropout(x, training=self.training)
         x = self.fc2(x)
-        return F.log_softmax(x, dim=1)
+        x = F.log_softmax(x, dim=1)
+        return x

Tutorials/06_pytorch_introduction/requirements.txt

@@ -0,0 +1,5 @@
+numpy
+opencv-python
+tqdm
+torch
+torchsummary

Tutorials/06_pytorch_introduction/test.py

@@ -6,7 +6,7 @@
 
 from model import Net
 
-path='Datasets/data' # Path where to save the mnist dataset
+path='Datasets/mnist' # Path where to save the downloaded mnist dataset
 def fetch(url):
     if os.path.exists(path) is False:
         os.makedirs(path)
@@ -32,21 +32,22 @@ def fetch(url):
 network.load_state_dict(torch.load("Models/06_pytorch_introduction/model.pt"))
 network.eval() # set to evaluation mode
 
-
+# loop over test images
 for test_image, test_target in zip(test_data, test_targets):
 
-    # convert to tensor
+    # normalize image and convert to tensor
     inference_image = torch.from_numpy(test_image).float() / 255.0
     inference_image = inference_image.unsqueeze(0).unsqueeze(0)
 
     # predict
     output = network(inference_image)
     pred = output.argmax(dim=1, keepdim=True)
+    prediction = str(pred.item())
 
     test_image = cv2.resize(test_image, (400, 400))
-    cv2.imshow(str(pred.item()), test_image)
+    cv2.imshow(prediction, test_image)
     key = cv2.waitKey(0)
     if key == ord('q'): # break on q key
         break
-
-cv2.destroyAllWindows()
+ 
+    cv2.destroyAllWindows()

Tutorials/06_pytorch_introduction/train.py

@@ -5,12 +5,14 @@
 import requests, gzip, os, hashlib
 
 import torch
-import torch.nn.functional as F
+import torch.nn as nn
 import torch.optim as optim
+from torchsummary import summary
 
 from model import Net
 
-path='Datasets/data'
+# define path to store dataset
+path='Datasets/mnist'
 def fetch(url):
     if os.path.exists(path) is False:
         os.makedirs(path)
@@ -33,12 +35,14 @@ def fetch(url):
 
 # uncomment to show images from dataset using OpenCV
 # for train_image, train_target in zip(train_data, train_targets):
-#     train_image = cv2.resize(train_image, (300, 300))
+#     train_image = cv2.resize(train_image, (400, 400))
 #     cv2.imshow("Image", train_image)
-#     cv2.waitKey(0)
-#     cv2.destroyAllWindows()
+#     # if Q button break this loop
+#     if cv2.waitKey(0) & 0xFF == ord('q'):
+#         break
+# cv2.destroyAllWindows()
 
-# define hyperparameters
+# define training hyperparameters
 n_epochs = 5
 batch_size_train = 64
 batch_size_test = 64
@@ -56,50 +60,67 @@ def fetch(url):
 test_batches = [np.array(test_data[i:i+batch_size_test]) for i in range(0, len(test_data), batch_size_test)]
 test_target_batches = [np.array(test_targets[i:i+batch_size_test]) for i in range(0, len(test_targets), batch_size_test)]
 
-# create network and optimizer
+# create network
 network = Net()
+
+# uncomment to print network summary
+summary(network, (1, 28, 28), device="cpu")
+
+# define loss function and optimizer
 optimizer = optim.Adam(network.parameters(), lr=learning_rate)
+loss_function = nn.CrossEntropyLoss()
 
 # create training loop
 def train(epoch):
+    # set network to training mode
     network.train()
 
     loss_sum = 0
+    # create a progress bar
     train_pbar = tqdm(zip(train_batches, train_target_batches), total=len(train_batches))
-    for data, target in train_pbar:
+    for index, (data, target) in enumerate(train_pbar, start=1):
 
         # convert data to torch.FloatTensor
         data = torch.from_numpy(data).float()
         target = torch.from_numpy(target).long()
 
+        # zero the parameter gradients
         optimizer.zero_grad()
+
+        # forward + backward + optimize
         output = network(data)
-        loss = F.nll_loss(output, target)
+        loss = loss_function(output, target)
         loss.backward()
         optimizer.step()
 
+        # update progress bar with loss value
         loss_sum += loss.item()
-        train_pbar.set_description(f"Epoch {epoch}, loss: {loss_sum / len(train_batches):.4f}")
+        train_pbar.set_description(f"Epoch {epoch}, loss: {loss_sum / index:.4f}")
 
 # create testing loop
 def test(epoch):
+    # set network to evaluation mode
     network.eval()
 
-    correct = 0
-    loss_sum = 0
+    correct, loss_sum = 0, 0
+    # create progress bar
     val_pbar = tqdm(zip(test_batches, test_target_batches), total=len(test_batches))
     with torch.no_grad():
-        for data, target in val_pbar:
+        for index, (data, target) in enumerate(val_pbar, start=1):
+
             # convert data to torch.FloatTensor
             data = torch.from_numpy(data).float()
             target = torch.from_numpy(target).long()
 
+            # forward pass
             output = network(data)
-            loss_sum += F.nll_loss(output, target).item() / target.size(0)
+
+            # update progress bar with loss and accuracy values
+            loss_sum += loss_function(output, target).item() / target.size(0)
             pred = output.data.max(1, keepdim=True)[1]
             correct += pred.eq(target.data.view_as(pred)).sum() / target.size(0)
 
-            val_pbar.set_description(f"val_loss: {loss_sum / len(test_batches):.4f}, val_accuracy: {correct / len(test_batches):.4f}")
+            val_pbar.set_description(f"val_loss: {loss_sum / index:.4f}, val_accuracy: {correct / index:.4f}")
 
 # train and test the model
 for epoch in range(1, n_epochs + 1):

mltu/tensorflow/requirements.txt

@@ -1 +1,2 @@
-tensorflow<=2.10.1
+tensorflow<=2.10.1
+tf2onnx

requirements.txt

@@ -4,7 +4,6 @@ tqdm
 pandas
 numpy
 opencv-python
-tf2onnx
 onnxruntime # onnxruntime-gpu for GPU support
 librosa==0.9.2
 importlib==1.0.4

setup.py

@@ -41,7 +41,7 @@ def get_version(initpath: str) -> str:
         'gpu': ['onnxruntime-gpu'],
     },
     python_requires='>=3',
-    packages = ['mltu'],
+    packages = ['mltu', 'mltu.utils', 'mltu.torch', 'mltu.tensorflow'],
     include_package_data=True,
     project_urls={
         'Source': 'https://github.com/pythonlessons/mltu/',