This workshop is intended for an audience that is new to Keras with a very basic knowledge of Deep Learning.
Workshop Agenda:
Section 1:
- Introduction to Keras
- How to build and train a model using Keras APIs
- Using Callbacks
Section 2
- Building a CNN model using Keras Layers
- Data augmentation with ImageGenerator
- Feature extraction with pre-trained CNN model
- Introduciton to Fine Tuning
Section 3
- Word Embeddings in Keras
- Using pre-trained word embeddings with the Embedding layer
- Introduction to RNN layers
Section 4
- Workflow for solving Text Classification problems
- Using a N-gram model
- Using a Sequence model
Installation instructions
Required packages:
-
Install Anaconda and create an environment that you can use for this workshop.
-
Install TensorFlow. There is a section specific to installing TensorFlow in a conda environment.
-
Install Keras.
-
Install Pillow.
-
Confirm that TensorBoard is installed.
Required Datasets:
-
Download the Cats Vs Dogs dataset from the Kaggle website. You will have to create an account if you don't have one already.
-
Download the GloVe pre-trained embedding from the GloVe website.
-
Download the IMDB dataset from this website
-
Download the Rotten Tomatotes dataset from the Kaggle site
Verify Installation:
-
git clone https://github.com/anj-s/kdd2018.git
-
Activate your conda environment
-
Start the Jupyter notebook by running "jupyter notebook"
-
Verify that you can run Keras and TensorFlow by running the "Verify Installation" jupyter notebook.
NOTE: You will need to modify the data directory paths in the notebooks since you will be pointing to a local directory.