Conv Nets TF

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
import numpy as np
# Self-contained course computer-vision helpers.
from pathlib import Path
import glob
from PIL import Image

def plot_convolution(data, kernel, title=""):
    fig, axes = plt.subplots(2, len(data) + 1, figsize=(10, 4))
    fig.suptitle(title, fontsize=16)
    tensor_kernel = np.asarray(kernel, dtype=np.float32)[:, :, None, None]
    for index, image in enumerate(data):
        axes[0, index].imshow(image, cmap="gray")
        tensor_image = np.asarray(image, dtype=np.float32)[None, :, :, None]
        convolved = tf.nn.conv2d(tensor_image, tensor_kernel, strides=1, padding="SAME")
        axes[1, index].imshow(convolved[0, :, :, 0], cmap="gray")
        axes[0, index].axis("off")
        axes[1, index].axis("off")
    axes[0, -1].imshow(kernel, cmap="gray")
    axes[0, -1].axis("off")
    axes[1, -1].axis("off")
    plt.tight_layout()
    plt.show()

def plot_results(history):
    fig, axes = plt.subplots(1, 2, figsize=(12, 4))
    axes[0].set_title("Accuracy")
    axes[1].set_title("Loss")
    for key in ("accuracy", "val_accuracy"):
        if key in history.history:
            axes[0].plot(history.history[key], label=key)
    for key in ("loss", "val_loss"):
        if key in history.history:
            axes[1].plot(history.history[key], label=key)
    axes[0].legend()
    axes[1].legend()
    plt.show()

def display_dataset(dataset, labels=None, n=10, classes=None):
    count = min(n, len(dataset))
    fig, axes = plt.subplots(1, count, figsize=(1.8 * count, 3))
    axes = np.atleast_1d(axes)
    for index in range(count):
        axes[index].imshow(dataset[index])
        axes[index].axis("off")
        if classes is not None and labels is not None:
            label = int(np.asarray(labels[index]).reshape(-1)[0])
            axes[index].set_title(classes[label])
    plt.show()

def check_image(filename):
    try:
        with Image.open(filename) as image:
            image.verify()
        return True
    except (OSError, SyntaxError):
        return False

def check_image_dir(pattern):
    invalid = [filename for filename in glob.glob(pattern) if not check_image(filename)]
    if invalid:
        names = ", ".join(Path(filename).name for filename in invalid[:5])
        raise ValueError(f"Invalid images detected ({len(invalid)}): {names}")
    return 0

# Bundled, deterministic handwritten-digits fixture.
from sklearn.datasets import load_digits as _course_load_digits

def _course_digits_load_data():
    course_digits = _course_load_digits()
    images = np.kron(
        course_digits.images.astype(np.float32),
        np.ones((3, 3), dtype=np.float32),
    )
    images = np.pad(images, ((0, 0), (2, 2), (2, 2))) * (255.0 / 16.0)
    labels = np.asarray(course_digits.target, dtype=np.int64)
    split = int(0.8 * len(images))
    return (images[:split], labels[:split]), (images[split:], labels[split:])

(x_train,y_train),(x_test,y_test) = _course_digits_load_data()
x_train = x_train.astype(np.float32) / 255.0
x_test = x_test.astype(np.float32) / 255.0

plot_convolution(x_train[:5],[[-1.,0.,1.],[-1.,0.,1.],[-1.,0.,1.]],'Vertical edge filter')
plot_convolution(x_train[:5],[[-1.,-1.,-1.],[0.,0.,0.],[1.,1.,1.]],'Horizontal edge filter')

model = keras.models.Sequential([
    keras.layers.Conv2D(filters=9, kernel_size=(5,5), input_shape=(28,28,1),activation='relu'),
    keras.layers.Flatten(),
    keras.layers.Dense(10)
])

model.compile(loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),metrics=['acc'])

model.summary()

x_train_c = np.expand_dims(x_train,3)
x_test_c = np.expand_dims(x_test,3)
hist = model.fit(x_train_c,y_train,validation_data=(x_test_c,y_test),epochs=2)

plot_results(hist)

fig,ax = plt.subplots(1,9)
l = model.layers[0].weights[0]
for i in range(9):
    ax[i].imshow(l[...,0,i])
    ax[i].axis('off')

model = keras.models.Sequential([
    keras.layers.Conv2D(filters=10, kernel_size=(5,5), input_shape=(28,28,1),activation='relu'),
    keras.layers.MaxPooling2D(),
    keras.layers.Conv2D(filters=20, kernel_size=(5,5), activation='relu'),
    keras.layers.MaxPooling2D(),
    keras.layers.Flatten(),
    keras.layers.Dense(10)
])

model.compile(loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),metrics=['acc'])

model.summary()

hist = model.fit(x_train_c,y_train,validation_data=(x_test_c,y_test),epochs=2)

plot_results(hist)

# Generated RGB patterns for deterministic, rights-safe execution.
def _course_rgb_load_data():
    yy, xx = np.mgrid[:32, :32]
    images, labels = [], []
    for label in range(10):
        base = np.zeros((32, 32, 3), dtype=np.float32)
        base[..., label % 3] = ((xx * (label + 2) + yy * ((label % 4) + 1)) % 32) / 31.0
        base[..., (label + 1) % 3] = ((yy + label * 3) % 32) / 31.0
        for sample in range(24):
            images.append(np.roll(base, shift=(sample % 5) - 2, axis=sample % 2))
            labels.append(label)
    images = (255.0 * np.asarray(images)).astype(np.float32)
    labels = np.asarray(labels, dtype=np.int64).reshape(-1, 1)
    split = 200
    return (images[:split], labels[:split]), (images[split:], labels[split:])

(x_train,y_train),(x_test,y_test) = _course_rgb_load_data()
x_train = x_train.astype(np.float32) / 255.0
x_test = x_test.astype(np.float32) / 255.0
classes = ('plane', 'car', 'bird', 'cat',
           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')

display_dataset(x_train,y_train,classes=classes)

model = keras.models.Sequential([
    keras.layers.Conv2D(filters = 6, kernel_size = 5, strides = 1, activation = 'relu', input_shape = (32,32,3)),
    keras.layers.MaxPooling2D(pool_size = 2, strides = 2),
    keras.layers.Conv2D(filters = 16, kernel_size = 5, strides = 1, activation = 'relu'),
    keras.layers.MaxPooling2D(pool_size = 2, strides = 2),
    keras.layers.Flatten(),
    keras.layers.Dense(120, activation = 'relu'),
    keras.layers.Dense(84, activation = 'relu'),
    keras.layers.Dense(10, activation = 'softmax')])

model.summary()

model.compile(optimizer = 'adam', loss = 'sparse_categorical_crossentropy', metrics = ['acc'])
hist = model.fit(x_train,y_train,validation_data=(x_test,y_test),epochs=2)

plot_results(hist)