integrating neural network with project

Readme.md

@@ -35,14 +35,14 @@ VERTICAL_NUM_OF_FIELDS = 3
 HORIZONTAL_NUM_OF_FIELDS = 3
 ```
 \
-#####4.1 Save generated map:
+####4.1 Save generated map:
 ```bash
 python main.py --save-map
 ```
 Map will be saved in maps directory. 
 Generated filename: map-uuid
 
-#####4.2 Load map
+####4.2 Load map
 ```bash
 python main.py --load-map=name_of_map
 ```

app/__init__.py

@@ -42,21 +42,21 @@ class App:
     def keys_pressed_handler(self):
         keys = pygame.key.get_pressed()
 
-        if keys[pygame.K_m]:
-            self.__tractor.move()
-            print(self.__tractor)
-
         if keys[pygame.K_w]:
             self.__tractor.move()
             print(self.__tractor)
 
+        if keys[pygame.K_n]:
+            self.__bot_is_running.set()
+            self.__tractor.harvest_checked_fields_handler(self.__bot_is_running)
+
         if keys[pygame.K_h]:
             self.__tractor.harvest()
 
         if keys[pygame.K_v]:
             self.__tractor.sow()
 
-        if keys[pygame.K_n]:
+        if keys[pygame.K_j]:
             self.__tractor.hydrate()
 
         if keys[pygame.K_f]:

app/base_field.py

@@ -7,6 +7,9 @@ class BaseField:
     def __init__(self, img_path: str):
         self._img_path = img_path
 
+    def get_img_path(self):
+        return self._img_path
+
     def draw_field(self, screen: pygame.Surface, pos_x: int,
                    pos_y: int, is_centered: bool = False,
                    size: tuple = None, angle: float = 0.0) -> None:

app/neural_network.py

@@ -0,0 +1,113 @@
+#!/usr/bin/python3
+import os
+
+from tensorflow.keras.models import Sequential, save_model, load_model
+from tensorflow.keras.layers import Dense, Flatten, Conv2D
+from tensorflow.keras.losses import sparse_categorical_crossentropy
+from tensorflow.keras.optimizers import Adam
+from tensorflow.keras.preprocessing.image import ImageDataGenerator
+from tensorflow import keras as k
+import numpy as np
+
+from app.base_field import BaseField
+from config import *
+
+
+class NeuralNetwork:
+    def __init__(self):
+        # Model config
+        self.batch_size = 25
+        self.img_width, self.img_height, self.img_num_channels = 25, 25, 3
+        self.loss_function = sparse_categorical_crossentropy
+        self.no_classes = 7
+        self.no_epochs = 40
+        self.optimizer = Adam()
+        self.verbosity = 1
+
+        # Determine shape of the data
+        self.input_shape = (self.img_width, self.img_height, self.img_num_channels)
+
+        # labels
+        self.labels = ["cabbage", "carrot", "corn", "lettuce", "paprika", "potato", "tomato"]
+
+    def init_model(self):
+        if not self.model_dir_is_empty():
+            # Load the model
+            self.model = load_model(
+                os.path.join(RESOURCE_DIR, "saved_model"),
+                custom_objects=None,
+                compile=True
+            )
+        else:
+            # Create the model
+            self.model = Sequential()
+            self.model.add(Conv2D(16, kernel_size=(5, 5), activation='relu', input_shape=self.input_shape))
+            self.model.add(Conv2D(32, kernel_size=(5, 5), activation='relu'))
+            self.model.add(Conv2D(64, kernel_size=(5, 5), activation='relu'))
+            self.model.add(Conv2D(128, kernel_size=(5, 5), activation='relu'))
+            self.model.add(Flatten())
+            self.model.add(Dense(16, activation='relu'))
+            self.model.add(Dense(self.no_classes, activation='softmax'))
+
+        # Display a model summary
+        self.model.summary()
+
+    def load_images(self):
+        # Create a generator
+        self.train_datagen = ImageDataGenerator(
+            rescale=1. / 255
+        )
+        self.train_datagen = self.train_datagen.flow_from_directory(
+            TRAINING_SET_DIR,
+            save_to_dir=ADAPTED_IMG_DIR,
+            save_format='jpeg',
+            batch_size=self.batch_size,
+            target_size=(25, 25),
+            class_mode='sparse')
+
+    def train(self):
+        self.model.compile(loss=self.loss_function,
+                           optimizer=self.optimizer,
+                           metrics=['accuracy'])
+
+        # Start training
+        self.model.fit(
+            self.train_datagen,
+            epochs=self.no_epochs,
+            shuffle=False)
+
+    def predict(self, field: BaseField) -> str:
+        print(field.get_img_path())
+        # corn_img_path = os.path.join(RESOURCE_DIR,'corn.png')
+        loaded_image = k.preprocessing.image.load_img(field.get_img_path(),
+                                                      target_size=(
+                                                          self.img_width, self.img_height, self.img_num_channels))
+
+        # convert to array and resample dividing by 255
+        img_array = k.preprocessing.image.img_to_array(loaded_image) / 255.
+
+        # add sample dimension. the predictor is expecting (1, CHANNELS, IMG_WIDTH, IMG_HEIGHT)
+        img_np_array = np.expand_dims(img_array, axis=0)
+        # print(img_np_array)
+        predictions = self.model.predict(img_np_array)
+        prediction = np.argmax(predictions[0])
+
+        label = self.labels[prediction]
+        print(f'Ground truth: {type(field).__name__} - Prediction: {label}')
+        return label
+
+    def model_dir_is_empty(self) -> bool:
+        if len(os.listdir(MODEL_DIR)) == 0:
+            return True
+        return False
+
+    def check(self, field: BaseField) -> str:
+        self.load_images()
+        self.init_model()
+        prediction = self.predict(field)
+
+        # Saving model
+        if not self.model_dir_is_empty():
+            save_model(self.model, MODEL_DIR)
+
+        return prediction

app/tractor.py

@@ -11,6 +11,7 @@ from typing import Union
 
 from app.base_field import BaseField
 from app.board import Board
+from app.neural_network import NeuralNetwork
 from app.utils import get_class
 from app.fields import CROPS, PLANTS, Crops, Sand, Clay, Field
 from config import *
@@ -28,6 +29,7 @@ class Tractor(BaseField):
         self.__board = board
         self.__harvested_corps = []
         self.__fuel = 10
+        self.__neural_network = None
 
     def draw(self, screen: pygame.Surface) -> None:
         self.draw_field(screen, self.__pos_x + FIELD_SIZE / 2, self.__pos_y + FIELD_SIZE / 2,
@@ -185,7 +187,7 @@ class Tractor(BaseField):
 
     def run_bot(self, moves: list[tuple[str, str]], is_running: threading.Event) -> None:
         print(moves)
-        print(f"Length of Moves {len(moves)}") #- {3 ** len(moves)}")
+        print(f"Length of Moves {len(moves)}")  # - {3 ** len(moves)}")
         while len(moves) > 0:
             movement, action = moves.pop(0)
             # do action
@@ -203,6 +205,12 @@ class Tractor(BaseField):
             time.sleep(1)
 
             # move
+            self.move_or_rotate(movement)
+
+            time.sleep(TIME_OF_MOVING)
+        is_running.clear()
+
+    def move_or_rotate(self, movement: str):
         print(f"Move {movement}")
         if movement == M_GO_FORWARD:
             self.move()
@@ -211,9 +219,6 @@ class Tractor(BaseField):
         elif movement == M_ROTATE_RIGHT:
             self.rotate_right()
 
-            time.sleep(TIME_OF_MOVING)
-        is_running.clear()
-
     @staticmethod
     def move_is_correct(x: int, y: int, direction: float) -> Union[(int, int), None]:
         pos_x = x * FIELD_SIZE
@@ -261,3 +266,26 @@ class Tractor(BaseField):
         obj = get_class("app.fields", choosen_type)
         board.get_fields()[x][y] = obj()
         return obj()
+
+    def harvest_checked_fields_handler(self, is_running: threading.Event):
+        thread = threading.Thread(target=self.harvest_checked_fields, args=(is_running,), daemon=True)
+        thread.start()
+
+    def harvest_checked_fields(self, is_running: threading.Event):
+        moves = [M_GO_FORWARD, M_ROTATE_LEFT, M_ROTATE_RIGHT]
+        distribution=[0.6,0.2,0.2]
+        while True:
+            field = self.get_field_from_board()
+
+            self.__neural_network = NeuralNetwork()
+            prediction = self.__neural_network.check(field)
+
+            if prediction.capitalize() in CROPS:
+                self.harvest()
+                break
+            chosen_move = random.choices(moves,distribution)
+
+            self.move_or_rotate(chosen_move[0])
+            time.sleep(1)
+
+        is_running.clear()

config.py

@@ -12,7 +12,8 @@ __all__ = (
     'A_SOW', 'A_HARVEST', 'A_HYDRATE', 'A_FERTILIZE', 'A_DO_NOTHING',
     'D_NORTH', 'D_EAST', 'D_SOUTH', 'D_WEST',
     'VALUE_OF_CROPS', 'VALUE_OF_PLANT', 'VALUE_OF_SAND', 'VALUE_OF_CLAY',
-    'MAP_FILE_NAME', 'JSON','SAVE_MAP', 'LOAD_MAP'
+    'MAP_FILE_NAME', 'JSON', 'SAVE_MAP', 'LOAD_MAP',
+    'TRAINING_SET_DIR', 'TEST_SET_DIR', 'ADAPTED_IMG_DIR', 'MODEL_DIR'
 )
 
 # Board settings:
@@ -26,11 +27,15 @@ HEIGHT = VERTICAL_NUM_OF_FIELDS * FIELD_SIZE
 FPS = 10
 CAPTION = 'Tractor'
 
-# Path
+# Paths
 BASE_DIR = os.path.dirname(__file__)
 RESOURCE_DIR = os.path.join(BASE_DIR, 'resources')
 MAP_DIR = os.path.join(BASE_DIR, 'maps')
 MAP_FILE_NAME = 'map'
+TRAINING_SET_DIR = os.path.join(RESOURCE_DIR, 'smaller_train')
+TEST_SET_DIR = os.path.join(RESOURCE_DIR, 'smaller_test')
+ADAPTED_IMG_DIR = os.path.join(RESOURCE_DIR, "adapted-images")
+MODEL_DIR = os.path.join(RESOURCE_DIR, 'saved_model')
 
 # Picture format
 PNG = "png"

resources/saved_model/keras_metadata.pb

@@ -9,5 +9,5 @@
 “root.layer-4"_tf_keras_layer*é{"name": "flatten", "trainable": true, "expects_training_arg": false, "dtype": "float32", "batch_input_shape": null, "stateful": false, "must_restore_from_config": false, "class_name": "Flatten", "config": {"name": "flatten", "trainable": true, "dtype": "float32", "data_format": "channels_last"}, "shared_object_id": 13, "input_spec": {"class_name": "InputSpec", "config": {"dtype": null, "shape": null, "ndim": null, "max_ndim": null, "min_ndim": 1, "axes": {}}, "shared_object_id": 26}}2


 Êroot.layer_with_weights-4"_tf_keras_layer*“{"name": "dense", "trainable": true, "expects_training_arg": false, "dtype": "float32", "batch_input_shape": null, "stateful": false, "must_restore_from_config": false, "class_name": "Dense", "config": {"name": "dense", "trainable": true, "dtype": "float32", "units": 16, "activation": "relu", "use_bias": true, "kernel_initializer": {"class_name": "GlorotUniform", "config": {"seed": null}, "shared_object_id": 14}, "bias_initializer": {"class_name": "Zeros", "config": {}, "shared_object_id": 15}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}, "shared_object_id": 16, "input_spec": {"class_name": "InputSpec", "config": {"dtype": null, "shape": null, "ndim": null, "max_ndim": null, "min_ndim": 2, "axes": {"-1": 10368}}, "shared_object_id": 27}, "build_input_shape": {"class_name": "TensorShape", "items": [null, 10368]}}2


 Êroot.layer_with_weights-5"_tf_keras_layer*“{"name": "dense_1", "trainable": true, "expects_training_arg": false, "dtype": "float32", "batch_input_shape": null, "stateful": false, "must_restore_from_config": false, "class_name": "Dense", "config": {"name": "dense_1", "trainable": true, "dtype": "float32", "units": 7, "activation": "softmax", "use_bias": true, "kernel_initializer": {"class_name": "GlorotUniform", "config": {"seed": null}, "shared_object_id": 17}, "bias_initializer": {"class_name": "Zeros", "config": {}, "shared_object_id": 18}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}, "shared_object_id": 19, "input_spec": {"class_name": "InputSpec", "config": {"dtype": null, "shape": null, "ndim": null, "max_ndim": null, "min_ndim": 2, "axes": {"-1": 16}}, "shared_object_id": 28}, "build_input_shape": {"class_name": "TensorShape", "items": [null, 16]}}2


-¹
croot.keras_api.metrics.0"_tf_keras_metric*‚
{"class_name": "Mean", "name": "loss", "dtype": "float32", "config": {"name": "loss", "dtype": "float32"}, "shared_object_id": 29}2


+¹
froot.keras_api.metrics.0"_tf_keras_metric*‚
{"class_name": "Mean", "name": "loss", "dtype": "float32", "config": {"name": "loss", "dtype": "float32"}, "shared_object_id": 29}2


-ó
droot.keras_api.metrics.1"_tf_keras_metric*¼
{"class_name": "MeanMetricWrapper", "name": "accuracy", "dtype": "float32", "config": {"name": "accuracy", "dtype": "float32", "fn": "sparse_categorical_accuracy"}, "shared_object_id": 22}2


+ó
groot.keras_api.metrics.1"_tf_keras_metric*¼
{"class_name": "MeanMetricWrapper", "name": "accuracy", "dtype": "float32", "config": {"name": "accuracy", "dtype": "float32", "fn": "sparse_categorical_accuracy"}, "shared_object_id": 22}2