Implement a Minimal Viable Data Product Using Deeplearning4J

This project demonstrates data flows and interaction between core components of a data-product. Our example use-case is automatic classification of images using neural networks.

The code examples show how:

• to use DL4J in a Spark session / CDSW
• persist labeled training data in Kudu (in order to slice and dice the training set)
• execute a learned model in a Spark-shell session
• execute a learned model in a Spark-streaming job

This means: we go from learning to production, not perfectly robust, but end-2-end!

Let’s go and implement the data product.

This is our todo-list:

1. Ingest and convert raw images
2. Train a model from labeled images
3. Query for a specific training set
4. Variation of model parameters
5. Evaluation of model quality
6. Predict the class of unknown images