Skip to content

Commit 5fe0983

Browse files
jElhammjElhamm
committed
Add Conclusions Code
1 parent fcc8154 commit 5fe0983

File tree

1 file changed

+252
-0
lines changed

1 file changed

+252
-0
lines changed

‎Chapter14/chapter_14.ipynb‎

Lines changed: 252 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,252 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "code",
5+
"source": [],
6+
"metadata": {
7+
"id": "4jVhBI-bTcHp"
8+
},
9+
"execution_count": null,
10+
"outputs": []
11+
},
12+
{
13+
"cell_type": "markdown",
14+
"metadata": {
15+
"id": "QmRAEtvJ_6OT"
16+
},
17+
"source": [
18+
"# **Deep Learning With Python - CHAPTER 14**"
19+
]
20+
},
21+
{
22+
"cell_type": "markdown",
23+
"metadata": {
24+
"id": "8HE2uMKK_7Qp"
25+
},
26+
"source": [
27+
"- This script provides a modular and well-structured implementation of various deep learning models using TensorFlow and Keras.\n",
28+
"\n",
29+
"- It defines separate classes for different model architectures, including **Dense-based**, **CNN-based**, **LSTM-based**, and **Transformer-based** models. Each class encapsulates the logic for building and compiling models with configurable parameters, making the code highly reusable and flexible.\n",
30+
"\n",
31+
"- The `DenseModel` class handles fully connected networks for classification and regression tasks, `CNNModel` builds convolutional networks for image processing, `LSTMModel` constructs recurrent networks for sequence data, and `TransformerModel` implements transformer-based models for NLP tasks.\n",
32+
"\n",
33+
"- This structured approach improves readability, maintainability, and scalability, making it easier to extend or modify individual components."
34+
]
35+
},
36+
{
37+
"cell_type": "code",
38+
"execution_count": 17,
39+
"metadata": {
40+
"id": "oqNg9frRCBgk"
41+
},
42+
"outputs": [],
43+
"source": [
44+
"import tensorflow as tf\n",
45+
"from tensorflow import keras\n",
46+
"from tensorflow.keras import layers"
47+
]
48+
},
49+
{
50+
"cell_type": "code",
51+
"execution_count": 18,
52+
"metadata": {
53+
"id": "7HHCQEtlu8Pg"
54+
},
55+
"outputs": [],
56+
"source": [
57+
"class DenseModel:\n",
58+
" \"\"\"Class for creating fully connected (Dense) models for different tasks.\"\"\"\n",
59+
" def __init__(self, input_shape, num_units=32):\n",
60+
" self.input_shape = input_shape\n",
61+
" self.num_units = num_units\n",
62+
"\n",
63+
" def build_binary_classification_model(self):\n",
64+
" inputs = keras.Input(shape=(self.input_shape,))\n",
65+
" x = layers.Dense(self.num_units, activation=\"relu\")(inputs)\n",
66+
" x = layers.Dense(self.num_units, activation=\"relu\")(x)\n",
67+
" outputs = layers.Dense(1, activation=\"sigmoid\")(x)\n",
68+
" model = keras.Model(inputs, outputs)\n",
69+
" model.compile(optimizer=\"rmsprop\", loss=\"binary_crossentropy\")\n",
70+
" return model\n",
71+
"\n",
72+
" def build_multiclass_classification_model(self, num_classes):\n",
73+
" inputs = keras.Input(shape=(self.input_shape,))\n",
74+
" x = layers.Dense(self.num_units, activation=\"relu\")(inputs)\n",
75+
" x = layers.Dense(self.num_units, activation=\"relu\")(x)\n",
76+
" outputs = layers.Dense(num_classes, activation=\"softmax\")(x)\n",
77+
" model = keras.Model(inputs, outputs)\n",
78+
" model.compile(optimizer=\"rmsprop\", loss=\"categorical_crossentropy\")\n",
79+
" return model\n",
80+
"\n",
81+
" def build_multilabel_classification_model(self, num_classes):\n",
82+
" inputs = keras.Input(shape=(self.input_shape,))\n",
83+
" x = layers.Dense(self.num_units, activation=\"relu\")(inputs)\n",
84+
" x = layers.Dense(self.num_units, activation=\"relu\")(x)\n",
85+
" outputs = layers.Dense(num_classes, activation=\"sigmoid\")(x)\n",
86+
" model = keras.Model(inputs, outputs)\n",
87+
" model.compile(optimizer=\"rmsprop\", loss=\"binary_crossentropy\")\n",
88+
" return model\n",
89+
"\n",
90+
" def build_regression_model(self, num_values):\n",
91+
" inputs = keras.Input(shape=(self.input_shape,))\n",
92+
" x = layers.Dense(self.num_units, activation=\"relu\")(inputs)\n",
93+
" x = layers.Dense(self.num_units, activation=\"relu\")(x)\n",
94+
" outputs = layers.Dense(num_values)(x)\n",
95+
" model = keras.Model(inputs, outputs)\n",
96+
" model.compile(optimizer=\"rmsprop\", loss=\"mse\")\n",
97+
" return model"
98+
]
99+
},
100+
{
101+
"cell_type": "code",
102+
"source": [
103+
"class CNNModel:\n",
104+
" \"\"\"Class for creating CNN-based models.\"\"\"\n",
105+
" def __init__(self, input_shape, num_classes):\n",
106+
" self.input_shape = input_shape\n",
107+
" self.num_classes = num_classes\n",
108+
"\n",
109+
" def build_model(self):\n",
110+
" inputs = keras.Input(shape=self.input_shape)\n",
111+
" x = layers.SeparableConv2D(32, 3, activation=\"relu\")(inputs)\n",
112+
" x = layers.SeparableConv2D(64, 3, activation=\"relu\")(x)\n",
113+
" x = layers.MaxPooling2D(2)(x)\n",
114+
" x = layers.SeparableConv2D(64, 3, activation=\"relu\")(x)\n",
115+
" x = layers.SeparableConv2D(128, 3, activation=\"relu\")(x)\n",
116+
" x = layers.MaxPooling2D(2)(x)\n",
117+
" x = layers.SeparableConv2D(64, 3, activation=\"relu\")(x)\n",
118+
" x = layers.SeparableConv2D(128, 3, activation=\"relu\")(x)\n",
119+
" x = layers.GlobalAveragePooling2D()(x)\n",
120+
" x = layers.Dense(32, activation=\"relu\")(x)\n",
121+
" outputs = layers.Dense(self.num_classes, activation=\"softmax\")(x)\n",
122+
" model = keras.Model(inputs, outputs)\n",
123+
" model.compile(optimizer=\"rmsprop\", loss=\"categorical_crossentropy\")\n",
124+
" return model"
125+
],
126+
"metadata": {
127+
"id": "6mitfoIAJY0r"
128+
},
129+
"execution_count": 19,
130+
"outputs": []
131+
},
132+
{
133+
"cell_type": "code",
134+
"source": [
135+
"class LSTMModel:\n",
136+
" \"\"\"Class for creating LSTM-based models for sequence processing.\"\"\"\n",
137+
" def __init__(self, input_shape, num_classes):\n",
138+
" self.input_shape = input_shape\n",
139+
" self.num_classes = num_classes\n",
140+
"\n",
141+
" def build_single_layer_lstm(self):\n",
142+
" inputs = keras.Input(shape=self.input_shape)\n",
143+
" x = layers.LSTM(32)(inputs)\n",
144+
" outputs = layers.Dense(self.num_classes, activation=\"sigmoid\")(x)\n",
145+
" model = keras.Model(inputs, outputs)\n",
146+
" model.compile(optimizer=\"rmsprop\", loss=\"binary_crossentropy\")\n",
147+
" return model\n",
148+
"\n",
149+
" def build_multi_layer_lstm(self):\n",
150+
" inputs = keras.Input(shape=self.input_shape)\n",
151+
" x = layers.LSTM(32, return_sequences=True)(inputs)\n",
152+
" x = layers.LSTM(32, return_sequences=True)(x)\n",
153+
" x = layers.LSTM(32)(x)\n",
154+
" outputs = layers.Dense(self.num_classes, activation=\"sigmoid\")(x)\n",
155+
" model = keras.Model(inputs, outputs)\n",
156+
" model.compile(optimizer=\"rmsprop\", loss=\"binary_crossentropy\")\n",
157+
" return model\n"
158+
],
159+
"metadata": {
160+
"id": "qr8jz4GeLVbT"
161+
},
162+
"execution_count": 20,
163+
"outputs": []
164+
},
165+
{
166+
"cell_type": "code",
167+
"source": [
168+
"class TransformerModel:\n",
169+
" \"\"\"Class for creating Transformer-based models.\"\"\"\n",
170+
" def __init__(self, sequence_length, vocab_size, embed_dim, dense_dim, num_heads):\n",
171+
" self.sequence_length = sequence_length\n",
172+
" self.vocab_size = vocab_size\n",
173+
" self.embed_dim = embed_dim\n",
174+
" self.dense_dim = dense_dim\n",
175+
" self.num_heads = num_heads\n",
176+
"\n",
177+
" def build_encoder_decoder_model(self):\n",
178+
" encoder_inputs = keras.Input(shape=(self.sequence_length,), dtype=\"int64\")\n",
179+
" x = PositionalEmbedding(self.sequence_length, self.vocab_size, self.embed_dim)(encoder_inputs)\n",
180+
" encoder_outputs = TransformerEncoder(self.embed_dim, self.dense_dim, self.num_heads)(x)\n",
181+
"\n",
182+
" decoder_inputs = keras.Input(shape=(None,), dtype=\"int64\")\n",
183+
" x = PositionalEmbedding(self.sequence_length, self.vocab_size, self.embed_dim)(decoder_inputs)\n",
184+
" x = TransformerDecoder(self.embed_dim, self.dense_dim, self.num_heads)(x, encoder_outputs)\n",
185+
" decoder_outputs = layers.Dense(self.vocab_size, activation=\"softmax\")(x)\n",
186+
"\n",
187+
" transformer = keras.Model([encoder_inputs, decoder_inputs], decoder_outputs)\n",
188+
" transformer.compile(optimizer=\"rmsprop\", loss=\"categorical_crossentropy\")\n",
189+
" return transformer\n",
190+
"\n",
191+
" def build_text_classification_model(self):\n",
192+
" inputs = keras.Input(shape=(self.sequence_length,), dtype=\"int64\")\n",
193+
" x = PositionalEmbedding(self.sequence_length, self.vocab_size, self.embed_dim)(inputs)\n",
194+
" x = TransformerEncoder(self.embed_dim, self.dense_dim, self.num_heads)(x)\n",
195+
" x = layers.GlobalMaxPooling1D()(x)\n",
196+
" outputs = layers.Dense(1, activation=\"sigmoid\")(x)\n",
197+
" model = keras.Model(inputs, outputs)\n",
198+
" model.compile(optimizer=\"rmsprop\", loss=\"binary_crossentropy\")\n",
199+
" return model"
200+
],
201+
"metadata": {
202+
"id": "5KU0n2MpLXsb"
203+
},
204+
"execution_count": 21,
205+
"outputs": []
206+
},
207+
{
208+
"cell_type": "code",
209+
"source": [
210+
"if __name__ == \"__main__\":\n",
211+
" num_input_features = 20\n",
212+
" num_classes = 5\n",
213+
" num_values = 1\n",
214+
" height, width, channels = 32, 32, 3\n",
215+
" num_timesteps, num_features = 10, 15\n",
216+
" sequence_length, vocab_size, embed_dim, dense_dim, num_heads = 100, 20000, 64, 256, 8\n",
217+
"\n",
218+
" dense_model = DenseModel(num_input_features)\n",
219+
" binary_model = dense_model.build_binary_classification_model()\n",
220+
" multi_class_model = dense_model.build_multiclass_classification_model(num_classes)\n",
221+
" regression_model = dense_model.build_regression_model(num_values)\n",
222+
"\n",
223+
" cnn_model = CNNModel((height, width, channels), num_classes).build_model()\n",
224+
" lstm_model = LSTMModel((num_timesteps, num_features), num_classes)\n",
225+
" single_lstm = lstm_model.build_single_layer_lstm()\n",
226+
" multi_lstm = lstm_model.build_multi_layer_lstm()\n",
227+
"\n",
228+
" transformer = TransformerModel(sequence_length, vocab_size, embed_dim, dense_dim, num_heads)\n",
229+
" transformer_text_model = transformer.build_text_classification_model()"
230+
],
231+
"metadata": {
232+
"id": "2Wr8ixy2TnVw"
233+
},
234+
"execution_count": null,
235+
"outputs": []
236+
}
237+
],
238+
"metadata": {
239+
"colab": {
240+
"provenance": []
241+
},
242+
"kernelspec": {
243+
"display_name": "Python 3",
244+
"name": "python3"
245+
},
246+
"language_info": {
247+
"name": "python"
248+
}
249+
},
250+
"nbformat": 4,
251+
"nbformat_minor": 0
252+
}

0 commit comments

Comments
 (0)