Easy-deep-learning-with-Keras

Updates

Nov 14, 2020

The source code is updated and can be run on TF2.0 & Google Colaboratory. Please check out the Jupyter Notebook (.ipynb) files! Thank you very much for your patience and support!

If you are unfamiliar with data preprocessing, first review NumPy & Pandas sections of Python for data analysis materials.

Materials in this repository are for educational purposes. Source code is written in Python 3.6+ & Keras ver 2.0+ (Using TensorFlow backend - For advanced topics, basic understanding of TensorFlow mechanics is necessary)

1. Multilayer Perceptrons

1) Basics of MLP

Regression tasks with MLP
Classification tasks with MLP

2) Advanced MLP - 1

Weight initialization schemes
Activation functions (Nonlinearity)
Batch normalization
Optimizers
Dropout
Ensemble of models

3) Advanced MLP - 2

Putting it altogether

2. Convolutional Neural Networks

1) Basic CNN

Basics of CNN architecture

2) Advanced CNN - 1

Getting deeper with CNNs

3) Advanced CNN - 2

CNN for sentence classification (imdb)

4) Using pretrained models

Importing models already trained on ImageNet dataset (keras.applications)

3. Recurrent Neural Networks

1) Basic RNN

Understanding RNN architecture
Vanilla RNN (SimpleRNN)
Stacked vanilla RNN
LSTM
Stacked LSTM

2) Advanced RNN - 1

Deep RNNs
Bidirectional RNNs
Deep bidirectional RNNs

3) Advanced RNN - 2

CNN-RNN

4) Advanced RNN - 3

CuDNN LSTM
CuDNN GRU

4. Unsupervised Learning

1) Autoencoders

Autoencoder basics
Convolutional autoencoder
Dimensionality reduction using autoencoder

5. ETC

0) Creating models

Sequential API
Model Functional API

1) Image processing

Importing images

2) Keras callbacks

ModelCheckpoint
EarlyStopping
ReduceLROnPlateau

3) Using GPUs

Make your training process faster with CUDA & CuDNN

4) Model selection

Cross validation
Grid search
Random search

5) Class weighted learning

Learning under class imbalance situations

6) Model weights

Getting model weights
Loading & saving model weights

6. Examples

1) Digit Recognition with RNN

Simple RNN model
Stacked RNN model
Bidirectional RNN model
Simple LSTM model
Stacked LSTM model
Bidirectional LSTM model
Simple GRU model
Stacked GRU model
Bidirectional GRU model

2) Fashion item classification with MLP

Simple MLP
Autoencoder + MLP (dimensionality reduction)

3) Question generation with seq2seq (using Quora dataset)

- Generating similar questions with seq2seq model

4) CNN for sentence classification

CNN-static implementation of Kim 2014 paper

5) Sentiment Analysis in Korean (using Naver Sentiment Movie Corpus)

Sentiment analysis with Logistic Regression (using sklearn & TF)
Sentiment analysis with RNN

7. Text Analytics

Section with emphasis on text data analytics

1) Text processing

2) Word embedding

3) CNNs for text data

1-D Convolution for text analysis
CNN for setnence classification (Kim 2014)
Dynamic CNN for sentence modeling (Kalchbrenner et al 2014)
CNN for text categorization (Johnson and Zhang 2014)

buomsoo-kim / easy-deep-learning-with-keras Goto Github PK