neural network with comments

main.py

@@ -1,7 +1,7 @@
 import pygame
 import sys
 import random
-from settings import screen_height, screen_width, SIZE, SPECIES, block_size, tile, road_coords, directions
+from settings import SIZE, directions, draw_lines_on_window
 from src.map import drawRoads, seedForFirstTime, return_fields_list, WORLD_MATRIX, get_type_by_position
 from src.Tractor import Tractor
 from src.bfs import Astar
@@ -9,7 +9,7 @@ from src.Plant import Plant
 from src.Field import Field
 import pickle
 import os
-from src.ID3 import make_decision
+from src.ID3 import action
 import torch
 import src.neural_networks as neural_networks
 
@@ -43,15 +43,13 @@ def recognize_plants(fields, destination):
     else:
         pred = 'none'
     print(pred)
-
+    return pred
-
 
 # pygame initialization
 pygame.init()
 clock = pygame.time.Clock()
-#pygame.mouse.set_visible(False)
 
-#GAME SCREEN
+# GAME SCREEN
 screen = pygame.display.set_mode(SIZE)
 pygame.display.set_caption("Traktor_interaktor")
 background = pygame.image.load("assets/farmland.jpg")
@@ -60,14 +58,7 @@ screen.fill((90,50,20))
 background.fill((90,50,20))
 background = drawRoads(background)
 
-for line in range(26):
+draw_lines_on_window(background)
-    pygame.draw.line(background, (0, 0, 0), (0, line * block_size), (936, line * block_size))
-    pygame.draw.line(background, (0, 0, 0), (line * block_size, 0), (line * block_size, screen_height))
-    
-pygame.draw.line(background, (0, 0, 0), (968, 285), (1336 , 285))
-pygame.draw.line(background, (0, 0, 0), (968, 649), (1336 , 649))
-pygame.draw.line(background, (0, 0, 0), (968, 285), (968, 649))
-pygame.draw.line(background, (0, 0, 0), (1336, 285), (1336, 649))
 
 #TRACTOR
 tractor = Tractor('oil','manual', 'fuel', 'fertilizer1', 20)
@@ -82,20 +73,18 @@ plant_group = pygame.sprite.Group()
 plant_group = seedForFirstTime()
 fields = return_fields_list()
 
-
-
 #
 tractor_move = pygame.USEREVENT + 1
 pygame.time.set_timer(tractor_move, 200)
 moves = []
 goal_astar = Astar()
-mx=random.randrange(0, 936, 36)
+mx = random.randrange(0, 936, 36)
-my=random.randrange(0, 936, 36)
+my = random.randrange(0, 936, 36)
 destination = (mx, my)
 print("Destination: ", destination)
-mx=int((mx+18)/36)
+mx = int((mx+18)/36)
-my=int((my+18)/36)
+my = int((my+18)/36)
-print("Destination: ", mx,my)
+print("Destination: ", mx, my)
 
 #ID3 TREE LOADING
 dtree = pickle.load(open(os.path.join('src','tree.plk'),'rb'))
@@ -104,25 +93,6 @@ dtree = pickle.load(open(os.path.join('src','tree.plk'),'rb'))
 this_field = WORLD_MATRIX[mx][my]
 this_contain = Field.getContain(this_field)
 
-def action(this_contain):
-    if isinstance(this_contain, Plant): 
-        this_plant = this_contain
-        params=Plant.getParameters(this_plant)
-        # print(this_field)
-        #ID3 decision
-        decision=make_decision(params[0],params[1],params[2],params[3],params[4],tractor.fuel,tractor.capacity,params[5],dtree)
-        # print('wzorst',params[0],'wilgotnosc',params[1],'dni_od_nawiezienia',params[2],'pogoda',params[3],'zdrowa',params[4],'paliwo',tractor.fuel,'pojemnosc eq',tractor.capacity,'cena sprzedazy',params[5])
-        # print(decision)
-        if decision == 1:
-            print('Gotowe do zbioru')
-            return 1
-        else:
-            print('nie zbieramy')
-            return 0
-    else:
-        print('Road, no plant growing')
-        return 0
-
 
 moves = goal_astar.search(
     [tractor.rect.x, tractor.rect.y, directions[tractor.rotation]], destination)
@@ -142,7 +112,7 @@ if __name__ == "__main__":
             if event.type == pygame.KEYDOWN:
                 if event.key==pygame.K_RETURN:
                     tractor.collect(plant_group)
-                    recognize_plants(fields, destination)
+                    # recognize_plants(fields, destination)
                 if event.key == pygame.K_ESCAPE:
                     running = False
             if event.type == tractor_move:
@@ -151,11 +121,16 @@ if __name__ == "__main__":
                     step = moves_list.pop()  # pop the last element
                     moves = tuple(moves_list)  # convert back to tuple
                     tractor.movement(step[0])
-                    if tractor.rect.x == destination[0] and tractor.rect.y == destination[1] and action(this_contain) == 1:
+                    # checks if tractor is in destiantion field and make decision if it's ready to collect
+                    if tractor.rect.x == destination[0] and tractor.rect.y == destination[1] and action(this_contain, Plant, tractor, dtree) == 1:
+                        # show what should be in this field
                         print('expected:', expected_plant)
-                        if recognize_plants(fields, destination) == 'carrot' or 'potato' or 'wheat':
+                        # check if program correctly recognize plant
+                        if recognize_plants(fields, destination) == expected_plant:
+                            # if correctly  recognized than plant can be collected
                             tractor.collect(plant_group)
-                    
+                        else:
+                            print('wrong recognition')
                     
 
         Tractor.movement_using_keys(tractor)

settings.py

@@ -1,4 +1,5 @@
 from cmath import sqrt
+import pygame
 
 
 screen_width = 1368
@@ -19,3 +20,13 @@ field_size = field_width*field_height
 fields_amount = 25
 
 directions = {0: 'UP', 90: 'RIGHT', 180: 'DOWN', 270: 'LEFT'}
+
+def draw_lines_on_window(background):
+    for line in range(26):
+        pygame.draw.line(background, (0, 0, 0), (0, line * block_size), (936, line * block_size))
+        pygame.draw.line(background, (0, 0, 0), (line * block_size, 0), (line * block_size, screen_height))
+        
+    pygame.draw.line(background, (0, 0, 0), (968, 285), (1336 , 285))
+    pygame.draw.line(background, (0, 0, 0), (968, 649), (1336 , 649))
+    pygame.draw.line(background, (0, 0, 0), (968, 285), (968, 649))
+    pygame.draw.line(background, (0, 0, 0), (1336, 285), (1336, 649))

src/ID3.py

@@ -42,3 +42,22 @@ def learnTree():
 #                             #przy robaczywej == 1 daje ok czyli jak 1 to git jest mozna zbierac, ale planowalem inaczej
 # decision=make_decision(70,85,12,4,0,65,54,1500,dtree)
 # print(decision)
+
+def action(this_contain, Plant, tractor, dtree):
+    if isinstance(this_contain, Plant): 
+        this_plant = this_contain
+        params=Plant.getParameters(this_plant)
+        # print(this_field)
+        #ID3 decision
+        decision=make_decision(params[0],params[1],params[2],params[3],params[4],tractor.fuel,tractor.capacity,params[5],dtree)
+        # print('wzorst',params[0],'wilgotnosc',params[1],'dni_od_nawiezienia',params[2],'pogoda',params[3],'zdrowa',params[4],'paliwo',tractor.fuel,'pojemnosc eq',tractor.capacity,'cena sprzedazy',params[5])
+        # print(decision)
+        if decision == 1:
+            print('Gotowe do zbioru')
+            return 1
+        else:
+            print('nie zbieramy')
+            return 0
+    else:
+        print('Road, no plant growing')
+        return 0

src/neural_networks.py

@@ -6,24 +6,45 @@ from torch.optim import Adam
 from torch.autograd import Variable
 from torch.utils.data import DataLoader
 
+# Check if CUDA-enabled GPU is available and set the device
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+# Define the classes for classification
 classes = ['carrot', 'potato', 'wheat']
 
+# Set the paths for the training and test data directories
 train_path = 'assets/learning/train'
 test_path = 'assets/learning/test'
 
+#list of transforms to compose (lista przekształceń do utworzenia)
 transformer = torchvision.transforms.Compose([
+    # resize input image to the given size
     torchvision.transforms.Resize((150, 150)),
+    # convert image to tensor(muli dim array)
     torchvision.transforms.ToTensor(),
+    # normalize tensor image with wit mean and standard deviation
+    # normalize doesn't support PIL image -> that is why we do .ToTensor before
+    # output[channel] = (input[channel] - mean[channel]) / std[channel]
     torchvision.transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])
 ])
 
+
 class Net(nn.Module):
     def __init__(self, num_classes=3):
         super(Net, self).__init__()
+        # Sequential - ordered dictionary
+        # Define the convolutional layers
+        # The output of one layer serves as the input to the next layer (3->12->20->32)
         self.features = nn.Sequential(
+            # Applies a 2D convolution over an input signal composed of several input planes
+            # Stosuje splot 2D dla sygnału wejściowego złożonego z kilku płaszczyzn wejściowych
             nn.Conv2d(3, 12, kernel_size=3, stride=1, padding=1),
+            # parameter of torch.nn.BatchNorm2d is the number of dimensions/channels that output 
+            # from the last layer and come in to the batch norm layer.
             nn.BatchNorm2d(12),
+            # activation function relu(x) = { 0 if x<0, x if x > 0}
+            # after each layer, an activation function needs to be applied
+            # so as to make the network non-linear and fit complex data
             nn.ReLU(),
             nn.MaxPool2d(kernel_size=2),
             nn.Conv2d(12, 20, kernel_size=3, stride=1, padding=1),
@@ -32,72 +53,111 @@ class Net(nn.Module):
             nn.BatchNorm2d(32),
             nn.ReLU()
         )
+        # takes the flattened feature maps from the previous convolutional layers as input
         self.classifier = nn.Linear(32 * 75 * 75, num_classes)
 
     def forward(self, x):
+        # Forward pass through the network
         x = self.features(x)
+        # Pass the input through the sequential block of 
+        # convolutional layers and activation functions 
         x = x.view(x.size(0), -1)
+        # Reshape the tensor by flattening it along the second dimension        
         x = self.classifier(x)
+        # Pass the flattened tensor through the linear layer for classification
         return x
+        # Return the output tensor
 
 def train(dataloader, model, optimizer, loss_fn):
     model.train()
     size = len(dataloader.dataset)
+    # Get the total number of training examples
     for batch, (X, y) in enumerate(dataloader):
         X, y = X.to(device), y.to(device)
-
+        # Move the input tensors to the appropriate device (CPU or GPU)        
         optimizer.zero_grad()
+        # Clear the gradients of the model parameters        
         pred = model(X.float())
+        # Perform a forward pass to obtain the predicted outputs       
         loss = loss_fn(pred, y)
+        # Compute the loss between the predicted outputs and the ground truth labels        
         loss.backward()
+        # Perform backpropagation to compute the gradients of the model parameters        
         optimizer.step()
-
+        # Update the model parameters using the computed gradients        
         if batch % 5 == 0:
             current = batch * len(X)
+            # Compute the current batch size            
             print(f"loss: {loss.item():>7f}  [{current:>5d}/{size:>5d}]")
+            # Print the current loss and the progress of the training in material def accuracy
 
 def test(dataloader, model, loss_fn):
     model.eval()
+    # Set the model to evaluation mode    
     size = len(dataloader.dataset)
+    # Get the total number of examples in the dataloader    
     test_loss, correct = 0, 0
-
+    # Initialize variables to keep track of the total test 
+    # loss and the number of correct predictions   
     with torch.no_grad():
+        # Disable gradient computation        
         for X, y in dataloader:
             X, y = X.to(device), y.to(device)
+            # Move the input tensors to the appropriate device (CPU or GPU)            
             pred = model(X.float())
+            # Perform a forward pass to obtain the predicted outputs            
             test_loss += loss_fn(pred, y).item()
+            # Compute the loss between the predicted outputs and the ground truth labels            
             correct += (pred.argmax(1) == y).sum().item()
+            # Count the number of correct predictions
             
     test_loss /= size
+    # Calculate the average test loss    
     accuracy = 100.0 * correct / size
+    # Calculate the accuracy as a percentage    
     print(f"Test Error:\n Accuracy: {accuracy:.1f}%, Avg loss: {test_loss:.8f}\n")
+    # Print the test accuracy and average test loss
     
 def predict(img_path, model):
     image = Image.open(img_path).convert('RGB')
+    # Open the image file from the given path and convert it to RGB mode    
     image_tensor = transformer(image).unsqueeze(0).to(device)
+    # Apply the image transformation pipeline defined earlier and convert the image to a tensor
+    # Add an extra dimension at the beginning to represent the batch
+    # Move the image tensor to the appropriate device (CPU or GPU)    
     output = model(image_tensor)
+    # Pass the image tensor through the model to obtain the output logits    
     _, predicted_idx = torch.max(output, 1)
+    # Find the index of the predicted class by taking the maximum value along the second dimension    
     pred = classes[predicted_idx.item()]
+    # Retrieve the corresponding class label from the classes list using the predicted index    
     return pred
 
 def learn():
     num_epochs = 50
     batch_size = 64
-
+    # Create a dataset from the images in the train_path directory
     train_dataset = torchvision.datasets.ImageFolder(train_path, transform=transformer)
+    # Create a data loader for the train dataset to load data in batches
     train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
-
+    # Create a dataset from the images in the test_path directory
     test_dataset = torchvision.datasets.ImageFolder(test_path, transform=transformer)
+    # Create a data loader for the test dataset to load data in batches
     test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=True)
-
+    # Create an instance of the neural network model
     model = Net(len(classes)).to(device)
+    # Create an optimizer for updating the model parameters during training
     optimizer = Adam(model.parameters(), lr=1e-3, weight_decay=0.0001)
+    # Define the loss function for computing the training loss
     loss_fn = nn.CrossEntropyLoss()
-
+    # Perform training for the specified number of epochs
     for epoch in range(num_epochs):
         print(f"Epoch {epoch + 1}\n-------------------------------")        
+        # Train the model using the training data
         train(train_loader, model, optimizer, loss_fn)        
+        # Evaluate the model on the test data
         test(test_loader, model, loss_fn)
-
+    # Print a message indicating that the training is done
     print("Done!")
+    # Save the trained model's state dictionary to a file
     torch.save(model.state_dict(), 'plants2.model')