Deep Learning Guide

DashiBoard supports training and evaluating deep learning models via the StreamlinerCore package.

Streamliner cards can be inserted into any pipeline and receive / provide data to/from any other card.

Setup

As designing a deep learning model and devising a training procedure are complex tasks, DashiBoard request the user to pass in folders with example configuration files (see the static/model and static/training folders for examples).

When using a streamliner card, DashiBoard will

Visualization

The streamliner card comes with a default visualization (see Pipelines.visualize), displaying the trajectory of the loss function in the training and validation datasets.

Example pipeline

Below we see a typical deep learning pipeline, comprised of the following steps:

• data partition,
• data normalization,
• model training and evaluation.