ahjota / ai-accelerators

DataRobot AI Accelerators are repeatable, code-first workflows designed to help speed up model development, deployment and time to value using the DataRobot API.

Install the DataRobot Python Client Package.

Making a change to the README as a contribution workflow test.

🚀 Getting started

1. Clone this repo
2. Import the desired accelerator into your preferred notebook (e.g., jupyter, Kaggle, Databricks Notebooks, Google Colab). We recommend using DR-Notebooks.
3. Execute the notebook.
4. Learn and understand the accelerator content.
5. You should now be able to modify the accelerator to solve your own problem. The easiest place to start is to replace the input data with your own.

❔Support

Please report feedback and problems by opening a Github Issue in this repo. Please note: The code in these repos is sourced from the DataRobot user community and is not owned or maintained by DataRobot, Inc. You may need to make edits or updates for this code to function properly in your environment.

🤔 Finding the Accelerator you need

Title	Accelerator Topic	What it's good for	3rd Party Integration	Extensibility to other Integrations
Automated Feature Discovery with Multiple Tables	Automated Feature Discovery workflow	Repeatable end-to-end workflow for automated time-aware feature engineering in Snowflake from data to batch predictions	Snowflake	Horizontal approach that applies to classification/ regression problems with a time-component; applies to all datasources
Azure Storage End-to-End Workflow with DataRobot	Azure End-to-End	Repeatable end-to-end workflow in Azure from data to batch predictions	Azure	Low, Azure focused
Creating Custom Blueprints with Composable ML	Create custom blueprints	Learn how to create custom blueprints	-	High, advanced use of DataRobot API
Customizing Lift Charts	Custom lift charts	Learn how to create a custom lift chart	-	High, advanced use of DataRobot API
Deploying a DataRobot model into AWS SageMaker	AWS SageMaker	Build an AI/ML model within DataRobot which will then be deployed and hosted within AWS SageMaker	AWS	Low, AWS focused
End-to-end demand forecasting and retraining workflow	Workflow for setting up retraining for time series, and executing it when model degrades	Repeatable workflow for setting up retraining. Talks through nuances of tracking actuals in TS, failed models, and performance after retraining	Snowflake	High, time-series focused, could easily swap Snowflake for another data source
End-to-end Time Series Cold Start Demand Forecasting	Time series Demand Forecasting (cold start)	Timeseries (cold start) workflow. Incomplete history for series and new series are very common in TS. This is a playbook for handling these challenges. Follows the	Snowflake	High, generalizes to all multi-series problems. Can use other data source/destination
End-to-end Time Series Demand Forecasting Workflow	Time series Demand Forecasting	This is a template for any time series project with the API. Precursor to cold-start notebook	Snowflake	High, easily can swap out Snowflake for another datasource.
End-to-end workflow with DataRobot and Snowflake	Snowflake End-to-End	Repeatable end-to-end workflow in Snowflake from data to batch predictions	Snowflake	Low, Snowflake focused
End-to-End workflows with DataRobot and AWS (S3 and Athena)	AWS End-to-End	Repeatable end-to-end workflow in AWS from data to batch predictions	AWS	Low, AWS focused
End-to-end workflows with DataRobot and Databricks	DBX End-to-End	Repeatable end-to-end workflow in DBX from data to batch predictions	Databricks	Low, DBX focused
Feature Reduction with FIRE	Feature Selection with FIRE	Learn how to use FIRE for feature selection	-	High, advanced use of DataRobot API
Google Cloud Platform & BigQuery: End-to-end AI/ML modeling	GCP End-to-End	Repeatable end-to-end workflow in GCP from data to batch predictions	GCP	Low, GCP focused
Gramian Angular Fields and High Frequency Data	Apply visualAI on high frequency data	Learn how to use Gramian Angular Fields to generate features from high frequency time series datasets like sensor readings	-	High, horizontal use case approach
How to Move a Model	Model Migration	Sample code to move a model between clusters	-	High, advanced use of DataRobot API
Hyperparameter Optimization and Gridsearch via Python API	Hyperparameter tuning	Learn how to use the API for hyperparameter tuning	-	High, advanced use of DataRobot API
Leveraging Hyperscaler API for Efficient Data Enrichment	Enriching training data with GCP APIs for NLP (sentiment, etc)	Extensible workflow to add new features for text, but the structure extends to the Vision API, etc	GCP	High, could apply to Azure/ AWS, other GCP APIs
Model Fine-Tuning with Eureqa	Fine tune Eureqa blueprints	Learn how to fine tune Eureqa blueprints via the API	-	High, advanced use of DataRobot API
Model Selection via Custom Metrics	Custom metrics	Learn how to rank models with custom metrics	-	High, advanced use of DataRobot API
Monitoring AWS Sagemaker models with DataRobot MLOps	Train and host a Sagemaker model that can be monitored in DR ML Ops	Repeatable workflow for applying ML Ops to Sagemaker models, building DataRobot compatible Sagemaker containers to host, train, and monitor models	AWS	Low, AWS focused, but broadly applies to Sagemaker deployments
Multi-model Analysis	Easily compare insights from models across different projects	Feature impact, model error, and partial dependence from models across N projects, for easy comparison. Inspired from our internal churn models	-	High, DR / Matplotlib/Seaborn focused
Native integration DataRobot and Snowflake Snowpark	Use Snowpark for feature eng, and deploy/monitor a PPS in Snowflake with ML Ops	Snowpark integration, Java scoring code deployment with PPS in Snowflake	Snowflake	Low, Snowflake focused
Prepare & Leverage Image Data for DataRobot with Databricks	Image data Databricks	Base64 conversion for images, project creation, and scoring pipeline	Databricks	High, broadly applicable to computer vision
Present Datarobot prediction insights with Streamlit	Streamlit prediction insights app	Streamlit prediction insights app that can be easily customized, helps transform prediction explanations	Streamlit	High, Streamlit baseline template
Recommendation Engine with DataRobot's MultiLabel	Multilabel recommendation engine	Learn how to create a recommendation engine that combines multilabel and AFD	-	High, horizontal use case approach
Retail - Predict Factory Quantities for New Products	Predicting demand for new products with very limited (annual) data. Regression, not TS	Multimodal, base64 conversion, detailed EDA and business rule analysis	-	High, horizontal use case approach
Spectrograms and Numerics for High Frequency Classification	Advanced Feature Eng. and Visual AI with Acoustic Data	Converting audio or high frequency sensor data into visual features and numeric features for analysis. Extends to machine failures, sensor readings, EM signals in general	-	High, horizontal use case approach
Tackling Churn with ML - Before Modelling	Churn with ML	Teach the problem framing and data management steps required before modeling begins	-	High, horizontal use case approach
Time Series Demand Forecasting What-If App	What-if-app for timeseries demand forecasting	Streamlit what-if app for timeseries demand forecasting	Streamlit	High, horizontal use case approach
Use MLFlow with DataRobot API for tracking experimentation	Repeatable experiments, meta-data tracking	Repeatable experiments that use MLflow to track key metrics across experiments e.g. model factories with multiple project settings, feature derivation windows	MLFlow	High
End-to-end workflow with DataRobot and SAP Hana	SAP End-to-End	Repeatable end-to-end workflow in SAP from data to batch predictions. Includes building out SAP HANA environment	SAP	Low, SAP Focused
The Self-Join Technique: Machine Learning with Panel Data	Advanced Feature Eng and Modeling with Panel data/ irregular time-series data	Repeatable workflow to use Automated Feature Discovery to derive time-based features by joining one table to itself multiple times on different panel dimensions. Highly useful in healthcare, manufacturing, B2B/B2C data...any data with repeat observations per subject	-	High - powerful, horizontal approach that applies to all industries
Add external data to time series forecasts with ReadySignal	Data enrichment with 3rd party data to improve time-series modeling	Repeatable workflow to add external data such as weather, economic conditions, census data, etc, which can provide lift to time-series forecasts	Ready Signal	High
Netlift/ Uplift Modeling	Determine markeing campaign impact	Modeling approach, feature engineering and technical considerations to measure campaign effectiveness on prospects	-	High - horizontal approach

Links to YouTube

Coming soon!

Useful Links

Coming soon!