Deep Learning with Databricks

Description

This course begins by covering the basics of neural networks and the tensorflow.keras API. We will then focus on using Spark to scale our models, including distributed training, hyperparameter tuning, and inference, and the meanwhile leveraging MLflow to track, version, and manage these models. You will apply model interpretability libraries to explain model predictions. Further, you will learn the concepts behind Convolutional Neural Networks (CNNs) and transfer learning, and apply them to solve image classification tasks. We will wrap up the course by covering Recurrent Neural Networks (RNNs) and attention-based models for natural language processing (NLP) applications.

Duration

2 full days or 4 half days

Objectives

• Build deep learning models using tensorflow.keras
• Tune hyperparameters at scale with HyperOpt and Spark
• Track, version, and manage experiments using MLflow
• Perform distributed inference at scale using pandas UDFs
• Scale and train distributed deep learning models using Horovod
• Apply model interpretability libraries, such as SHAP, to understand model predictions
• Use CNNs and transfer learning for image classification tasks
• Use RNNs, attention-based models, and transfer learning for NLP tasks

Prerequisites

• Intermediate experience with Python/pandas
• Experience building machine learning models
• Familiarity with Apache Spark

Logistics

• Zoom is our chosen online platform to deliver classes. Ensure you can access Zoom by clicking here.
• Some classes may also leverage Slack for classroom communication. Please test Slack by clicking here. If you are having difficulty connecting to Slack, please disconnect from your VPN.
• If your company laptop has firewall restrictions, we recommend that you use a personal laptop for the training.
• Please have one of these supported browsers installed.

Outline

Day 1

• Neural network and tf.keras fundamentals
• Improve models by adding data standardization, callbacks, checkpointing, etc.
• Track and version models with MLflow
• Distributed inference with pandas UDFs
• Distributed hyperparameter tuning with HyperOpt
• Distributed model training with Horovod

Day 2

• Distributed model training with Horovod and Petastorm
• Model interpretability with LIME and SHAP
• CNNs for image classification and transfer learning
• Deploy REST endpoint using MLflow Model Serving on Databricks
• Bucketing
• Optimization with Adaptive Query Execution (AQE)
• Textual embeddings, RNNs, attention-based models, and transfer learning for named entity recognition (NER)

If you have any questions, please refer to our Frequently Asked Questions page.