Explaining Deep Learning in DataRobot

This post was originally part of the DataRobot Community. Visit now to browse discussions and ask questions about DataRobot, AI Platform, data science, and more.

This article provides an overview of the many algorithms for modeling in DataRobot, from classical methods such as linear regression, and random forest, to the latest deep learning methods from TensorFlow and Keras. DataRobot employs Deep Learning algorithms for regression and classification for Automated Machine Learning, Visual AI, and Automated Time Series.

Automated Machine Learning

Here we have AutoML which includes algorithms from Keras and TensorFlow. It supports multimodal deep learning, which means you can include numeric, categorical, text, and image data, all in the same model.

Figure 1. AutoML Leaderboard

Figure 2. AutoML Deep Learning

DataRobot also allows you to tune all of these deep learning models just as you would any of the machine learning models inside DataRobot. You can tune the number of layers, neurons per layer, activation functions, normalization, dropout rates, cell types, and more.

Figure 3. Advanced Tuning

Figure 4. Advanced Tuning parameters

Here is a partial list of the deep learning algorithms inside DataRobot.

• Feedforward Neural Networks
• Deep Residual Networks
• Self-Normalizing Neural Networks
• Adaptive Learning Networks
• Attention-based text mining networks
• Variational AutoEncoders
• State-of-the-art CNN architectures for images
• Pretrained CNN architectures for images
• Self-normalizing residual networks
• FastText
• Word2Vec
• Neural Architecture Search (using Hyperband)
• Deep and Cross Networks
• Neural Factorization Machines
• AutoInt, short for Automatic Feature Interaction Learning

Visual AI

In Visual AI, DataRobot employs several pretrained CNN architectures, which allow you to start building models with just a few hundred images.

Figure 5. Visual AI Leaderboard

Let’s look at some DataRobot’s pretrained CNNs.

Figure 6. Visual AI pretrained CNNs

Automated Time Series

Automated Time Series offers deep learning models in the form of Deep Learning Regressors, MultiSeries networks, Sequence to Sequence models, LSTM & GRU RNNs, along with the latest deep learning algorithm, DeepAR.

Here we can see some of the deep learning options available for time series forecasting.

Figure 7. Automated Time Series Deep Learning

If you have any questions,  click Comment (below) and ask them now.

More information

AI Wiki – Deep Learning Algorithms

DataRobot Public Documentation – Visual AI reference.

About the author

Linda Haviland

Community Manager

Meet Linda Haviland