CS50AI - Traffic - Solutions

  If you have doubts refer this.Don't try to copy and answer.

Name: CS50's Introduction to Artificial Intelligence with Python

University: Harvard University

URL: https://cs50.harvard.edu/ai/2020/

My GitHub: https://github.com/Dinesh-Kumar-E

Automatic CS50 Submit Tool: https://github.com/Dinesh-Kumar-E/CS50-Submit

See: https://cs50.harvard.edu/college/2021/fall/syllabus/#academic-honesty

 

traffic.py

 import cv2
import numpy as np
import os
import sys
import tensorflow as tf

from sklearn.model_selection import train_test_split

EPOCHS = 10
IMG_WIDTH = 30
IMG_HEIGHT = 30
NUM_CATEGORIES = 43
TEST_SIZE = 0.4


def main():

    # Check command-line arguments
    if len(sys.argv) not in [2, 3]:
        sys.exit("Usage: python traffic.py data_directory [model.h5]")

    # Get image arrays and labels for all image files
    images, labels = load_data(sys.argv[1])

    # Split data into training and testing sets
    labels = tf.keras.utils.to_categorical(labels)
    x_train, x_test, y_train, y_test = train_test_split(
        np.array(images), np.array(labels), test_size=TEST_SIZE
    )

    x_train, x_test = x_train / 255.0, x_test / 255.0

    # Get a compiled neural network
    model = get_model()

    # Fit model on training data
    model.fit(x_train, y_train, epochs=EPOCHS)

    # Evaluate neural network performance
    model.evaluate(x_test,  y_test, verbose=2)

    # Save model to file
    if len(sys.argv) == 3:
        filename = sys.argv[2]
        model.save(filename)
        print(f"Model saved to {filename}.")


def load_data(data_dir):
    """
    Load image data from directory `data_dir`.
    Assume `data_dir` has one directory named after each category, numbered
    0 through NUM_CATEGORIES - 1. Inside each category directory will be some
    number of image files.
    Return tuple `(images, labels)`. `images` should be a list of all
    of the images in the data directory, where each image is formatted as a
    numpy ndarray with dimensions IMG_WIDTH x IMG_HEIGHT x 3. `labels` should
    be a list of integer labels, representing the categories for each of the
    corresponding `images`.
    """

    images = []
    labels = []

    # iterate through data set directories
    for directory in os.listdir(data_dir):
        # iterate through single image files
        print(f"Started loading files from {directory} directory")
        for file in os.listdir(os.path.join(data_dir, directory)):
            image = cv2.imread(os.path.join(data_dir, directory, file))
            resized = cv2.resize(image, (IMG_WIDTH, IMG_HEIGHT))
            images.append(resized)
            labels.append(int(directory))
        print(f"Ended loading files from {directory} directory")
    return images, labels


def get_model():
    """
    Returns a compiled convolutional neural network model. Assume that the
    `input_shape` of the first layer is `(IMG_WIDTH, IMG_HEIGHT, 3)`.
    The output layer should have `NUM_CATEGORIES` units, one for each category.
    """
    model = tf.keras.models.Sequential([
        # Convolutional layer. Learn 32 filters using a 3x3 kernel
        tf.keras.layers.Conv2D(
            32, (3, 3), activation="relu", input_shape=(IMG_WIDTH, IMG_HEIGHT, 3)
        ),

        # Max-pooling layer, using 2x2 pool size
        tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),

        tf.keras.layers.Conv2D(
            32, (3, 3), activation="relu", input_shape=(IMG_WIDTH, IMG_HEIGHT, 3)
        ),

        # Max-pooling layer, using 2x2 pool size
        tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),

        # Flatten units
        tf.keras.layers.Flatten(),

        # Add a hidden layer with dropout
        tf.keras.layers.Dense(128, activation="relu"),
        tf.keras.layers.Dropout(0.5),

        # Add an output layer with output units for all 10 digits
        tf.keras.layers.Dense(NUM_CATEGORIES, activation="softmax")
    ])

    model.compile(
        optimizer="adam",
        loss="categorical_crossentropy",
        metrics=["accuracy"]
    )

    model.summary()

    return model


if __name__ == "__main__":
    main()