Skip to content

Commit 03def3a

Browse files
mattdangerwmattdangerw
committed
Add third edition notebooks
1 parent a6a24ed commit 03def3a

File tree

37 files changed

+18461
-1332
lines changed

37 files changed

+18461
-1332
lines changed

‎chapter02_mathematical-building-blocks.ipynb

Lines changed: 201 additions & 232 deletions
Large diffs are not rendered by default.

‎chapter03_introduction-to-ml-frameworks.ipynb

Lines changed: 1761 additions & 0 deletions
Large diffs are not rendered by default.

‎chapter04_getting-started-with-neural-networks.ipynb

Lines changed: 305 additions & 438 deletions
Large diffs are not rendered by default.

‎chapter05_fundamentals-of-ml.ipynb

Lines changed: 422 additions & 249 deletions
Large diffs are not rendered by default.

‎chapter07_working-with-keras.ipynb

Lines changed: 843 additions & 413 deletions
Large diffs are not rendered by default.

‎chapter08_image-classification.ipynb

Lines changed: 1004 additions & 0 deletions
Large diffs are not rendered by default.

‎chapter09_convnet-architecture-best-practices.ipynb

Lines changed: 356 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,356 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {
6+
"colab_type": "text"
7+
},
8+
"source": [
9+
"This is a companion notebook for the book [Deep Learning with Python, Third Edition](TODO). For readability, it only contains runnable code blocks and section titles, and omits everything else in the book: text paragraphs, figures, and pseudocode.\n\n**If you want to be able to follow what's going on, I recommend reading the notebook side by side with your copy of the book.**"
10+
]
11+
},
12+
{
13+
"cell_type": "code",
14+
"execution_count": 0,
15+
"metadata": {
16+
"colab_type": "code"
17+
},
18+
"outputs": [],
19+
"source": [
20+
"!pip install keras-nightly --upgrade -q"
21+
]
22+
},
23+
{
24+
"cell_type": "code",
25+
"execution_count": 0,
26+
"metadata": {
27+
"colab_type": "code"
28+
},
29+
"outputs": [],
30+
"source": [
31+
"import os\n",
32+
"os.environ[\"KERAS_BACKEND\"] = \"jax\""
33+
]
34+
},
35+
{
36+
"cell_type": "markdown",
37+
"metadata": {
38+
"colab_type": "text"
39+
},
40+
"source": [
41+
"### Modularity, hierarchy, and reuse"
42+
]
43+
},
44+
{
45+
"cell_type": "markdown",
46+
"metadata": {
47+
"colab_type": "text"
48+
},
49+
"source": [
50+
"### Residual connections"
51+
]
52+
},
53+
{
54+
"cell_type": "code",
55+
"execution_count": 0,
56+
"metadata": {
57+
"colab_type": "code"
58+
},
59+
"outputs": [],
60+
"source": [
61+
"import keras\n",
62+
"from keras import layers\n",
63+
"\n",
64+
"inputs = keras.Input(shape=(32, 32, 3))\n",
65+
"x = layers.Conv2D(32, 3, activation=\"relu\")(inputs)\n",
66+
"residual = x\n",
67+
"x = layers.Conv2D(64, 3, activation=\"relu\", padding=\"same\")(x)\n",
68+
"residual = layers.Conv2D(64, 1)(residual)\n",
69+
"x = layers.add([x, residual])"
70+
]
71+
},
72+
{
73+
"cell_type": "code",
74+
"execution_count": 0,
75+
"metadata": {
76+
"colab_type": "code"
77+
},
78+
"outputs": [],
79+
"source": [
80+
"inputs = keras.Input(shape=(32, 32, 3))\n",
81+
"x = layers.Conv2D(32, 3, activation=\"relu\")(inputs)\n",
82+
"residual = x\n",
83+
"x = layers.Conv2D(64, 3, activation=\"relu\", padding=\"same\")(x)\n",
84+
"x = layers.MaxPooling2D(2, padding=\"same\")(x)\n",
85+
"residual = layers.Conv2D(64, 1, strides=2)(residual)\n",
86+
"x = layers.add([x, residual])"
87+
]
88+
},
89+
{
90+
"cell_type": "code",
91+
"execution_count": 0,
92+
"metadata": {
93+
"colab_type": "code"
94+
},
95+
"outputs": [],
96+
"source": [
97+
"inputs = keras.Input(shape=(32, 32, 3))\n",
98+
"x = layers.Rescaling(1.0 / 255)(inputs)\n",
99+
"\n",
100+
"def residual_block(x, filters, pooling=False):\n",
101+
" residual = x\n",
102+
" x = layers.Conv2D(filters, 3, activation=\"relu\", padding=\"same\")(x)\n",
103+
" x = layers.Conv2D(filters, 3, activation=\"relu\", padding=\"same\")(x)\n",
104+
" if pooling:\n",
105+
" x = layers.MaxPooling2D(2, padding=\"same\")(x)\n",
106+
" residual = layers.Conv2D(filters, 1, strides=2)(residual)\n",
107+
" elif filters != residual.shape[-1]:\n",
108+
" residual = layers.Conv2D(filters, 1)(residual)\n",
109+
" x = layers.add([x, residual])\n",
110+
" return x\n",
111+
"\n",
112+
"x = residual_block(x, filters=32, pooling=True)\n",
113+
"x = residual_block(x, filters=64, pooling=True)\n",
114+
"x = residual_block(x, filters=128, pooling=False)\n",
115+
"\n",
116+
"x = layers.GlobalAveragePooling2D()(x)\n",
117+
"outputs = layers.Dense(1, activation=\"sigmoid\")(x)\n",
118+
"model = keras.Model(inputs=inputs, outputs=outputs)"
119+
]
120+
},
121+
{
122+
"cell_type": "markdown",
123+
"metadata": {
124+
"colab_type": "text"
125+
},
126+
"source": [
127+
"### Batch normalization"
128+
]
129+
},
130+
{
131+
"cell_type": "markdown",
132+
"metadata": {
133+
"colab_type": "text"
134+
},
135+
"source": [
136+
"### Depthwise separable convolutions"
137+
]
138+
},
139+
{
140+
"cell_type": "markdown",
141+
"metadata": {
142+
"colab_type": "text"
143+
},
144+
"source": [
145+
"### Putting it together: a mini Xception-like model"
146+
]
147+
},
148+
{
149+
"cell_type": "code",
150+
"execution_count": 0,
151+
"metadata": {
152+
"colab_type": "code"
153+
},
154+
"outputs": [],
155+
"source": [
156+
"import kagglehub\n",
157+
"\n",
158+
"kagglehub.login()"
159+
]
160+
},
161+
{
162+
"cell_type": "code",
163+
"execution_count": 0,
164+
"metadata": {
165+
"colab_type": "code"
166+
},
167+
"outputs": [],
168+
"source": [
169+
"import zipfile\n",
170+
"\n",
171+
"download_path = kagglehub.competition_download(\"dogs-vs-cats\")\n",
172+
"\n",
173+
"with zipfile.ZipFile(download_path + \"/train.zip\", \"r\") as zip_ref:\n",
174+
" zip_ref.extractall(\".\")"
175+
]
176+
},
177+
{
178+
"cell_type": "code",
179+
"execution_count": 0,
180+
"metadata": {
181+
"colab_type": "code"
182+
},
183+
"outputs": [],
184+
"source": [
185+
"import os, shutil, pathlib\n",
186+
"from keras.utils import image_dataset_from_directory\n",
187+
"\n",
188+
"original_dir = pathlib.Path(\"train\")\n",
189+
"new_base_dir = pathlib.Path(\"dogs_vs_cats_small\")\n",
190+
"\n",
191+
"def make_subset(subset_name, start_index, end_index):\n",
192+
" for category in (\"cat\", \"dog\"):\n",
193+
" dir = new_base_dir / subset_name / category\n",
194+
" os.makedirs(dir)\n",
195+
" fnames = [f\"{category}.{i}.jpg\" for i in range(start_index, end_index)]\n",
196+
" for fname in fnames:\n",
197+
" shutil.copyfile(src=original_dir / fname, dst=dir / fname)\n",
198+
"\n",
199+
"make_subset(\"train\", start_index=0, end_index=1000)\n",
200+
"make_subset(\"validation\", start_index=1000, end_index=1500)\n",
201+
"make_subset(\"test\", start_index=1500, end_index=2500)\n",
202+
"\n",
203+
"batch_size = 64\n",
204+
"image_size = (180, 180)\n",
205+
"train_dataset = image_dataset_from_directory(\n",
206+
" new_base_dir / \"train\",\n",
207+
" image_size=image_size,\n",
208+
" batch_size=batch_size,\n",
209+
")\n",
210+
"validation_dataset = image_dataset_from_directory(\n",
211+
" new_base_dir / \"validation\",\n",
212+
" image_size=image_size,\n",
213+
" batch_size=batch_size,\n",
214+
")\n",
215+
"test_dataset = image_dataset_from_directory(\n",
216+
" new_base_dir / \"test\",\n",
217+
" image_size=image_size,\n",
218+
" batch_size=batch_size,\n",
219+
")"
220+
]
221+
},
222+
{
223+
"cell_type": "code",
224+
"execution_count": 0,
225+
"metadata": {
226+
"colab_type": "code"
227+
},
228+
"outputs": [],
229+
"source": [
230+
"import tensorflow as tf\n",
231+
"from keras import layers\n",
232+
"\n",
233+
"data_augmentation_layers = [\n",
234+
" layers.RandomFlip(\"horizontal\"),\n",
235+
" layers.RandomRotation(0.1),\n",
236+
" layers.RandomZoom(0.2),\n",
237+
"]\n",
238+
"\n",
239+
"def data_augmentation(images, targets):\n",
240+
" for layer in data_augmentation_layers:\n",
241+
" images = layer(images)\n",
242+
" return images, targets\n",
243+
"\n",
244+
"augmented_train_dataset = train_dataset.map(\n",
245+
" data_augmentation, num_parallel_calls=8\n",
246+
")\n",
247+
"augmented_train_dataset = augmented_train_dataset.prefetch(tf.data.AUTOTUNE)"
248+
]
249+
},
250+
{
251+
"cell_type": "code",
252+
"execution_count": 0,
253+
"metadata": {
254+
"colab_type": "code"
255+
},
256+
"outputs": [],
257+
"source": [
258+
"import keras\n",
259+
"\n",
260+
"inputs = keras.Input(shape=(180, 180, 3))\n",
261+
"x = layers.Rescaling(1.0 / 255)(inputs)\n",
262+
"x = layers.Conv2D(filters=32, kernel_size=5, use_bias=False)(x)\n",
263+
"\n",
264+
"for size in [32, 64, 128, 256, 512]:\n",
265+
" residual = x\n",
266+
"\n",
267+
" x = layers.BatchNormalization()(x)\n",
268+
" x = layers.Activation(\"relu\")(x)\n",
269+
" x = layers.SeparableConv2D(size, 3, padding=\"same\", use_bias=False)(x)\n",
270+
"\n",
271+
" x = layers.BatchNormalization()(x)\n",
272+
" x = layers.Activation(\"relu\")(x)\n",
273+
" x = layers.SeparableConv2D(size, 3, padding=\"same\", use_bias=False)(x)\n",
274+
"\n",
275+
" x = layers.MaxPooling2D(3, strides=2, padding=\"same\")(x)\n",
276+
"\n",
277+
" residual = layers.Conv2D(\n",
278+
" size, 1, strides=2, padding=\"same\", use_bias=False\n",
279+
" )(residual)\n",
280+
" x = layers.add([x, residual])\n",
281+
"\n",
282+
"x = layers.GlobalAveragePooling2D()(x)\n",
283+
"x = layers.Dropout(0.5)(x)\n",
284+
"outputs = layers.Dense(1, activation=\"sigmoid\")(x)\n",
285+
"model = keras.Model(inputs=inputs, outputs=outputs)"
286+
]
287+
},
288+
{
289+
"cell_type": "code",
290+
"execution_count": 0,
291+
"metadata": {
292+
"colab_type": "code"
293+
},
294+
"outputs": [],
295+
"source": [
296+
"model.compile(\n",
297+
" loss=\"binary_crossentropy\",\n",
298+
" optimizer=\"adam\",\n",
299+
" metrics=[\"accuracy\"],\n",
300+
")\n",
301+
"history = model.fit(\n",
302+
" augmented_train_dataset,\n",
303+
" epochs=100,\n",
304+
" validation_data=validation_dataset,\n",
305+
")"
306+
]
307+
},
308+
{
309+
"cell_type": "markdown",
310+
"metadata": {
311+
"colab_type": "text"
312+
},
313+
"source": [
314+
"### Beyond convolution: Vision Transformers"
315+
]
316+
},
317+
{
318+
"cell_type": "markdown",
319+
"metadata": {
320+
"colab_type": "text"
321+
},
322+
"source": [
323+
"## Chapter summary"
324+
]
325+
}
326+
],
327+
"metadata": {
328+
"accelerator": "GPU",
329+
"colab": {
330+
"collapsed_sections": [],
331+
"name": "chapter09_convnet-architecture-best-practices",
332+
"private_outputs": false,
333+
"provenance": [],
334+
"toc_visible": true
335+
},
336+
"kernelspec": {
337+
"display_name": "Python 3",
338+
"language": "python",
339+
"name": "python3"
340+
},
341+
"language_info": {
342+
"codemirror_mode": {
343+
"name": "ipython",
344+
"version": 3
345+
},
346+
"file_extension": ".py",
347+
"mimetype": "text/x-python",
348+
"name": "python",
349+
"nbconvert_exporter": "python",
350+
"pygments_lexer": "ipython3",
351+
"version": "3.10.0"
352+
}
353+
},
354+
"nbformat": 4,
355+
"nbformat_minor": 0
356+
}

0 commit comments

Comments
 (0)