hands-on machine learning using healthcare
TRANSCRIPT
Hands-on Machine Learning Using healthcare.ai
February 8, 2017
2
Health Catalyst Data Science Team
Levi Thatcher Mike Mastanduno Taylor Larsen Taylor Miller
3
Purpose of today’s chat
• Explain what healthcare.ai does
• Get you started in RStudio
• Talk about the roadmap
• Q&A
4
What does healthcare.ai do?
• Enables quick model creation and deployment
• Common pre-processing functions
• Proper algorithms for healthcare
• Suitable metrics for model evaluation
• Easy deployment for nightly prediction
5
Who is healthcare.ai for?
• Technical folks
• Business intelligence folks
• Data scientists
6
Poll question – Operating System
At work, what operating system would you use for R/Python? 510 respondents
• Windows – 79%• Mac – 11%• Linux – 11%
7
Poll question – R vs Python
Have you ever run R or Python code before? 526 respondents
• R – 22%• Python – 14%• Neither – 36%• Both – 28%
8
Algorithm choices for healthcare.ai
• Lasso
• Random Forest
9
Difference between R and Python packages?
• R package currently has more functionality
• Python will be used to leverage large datasets
*Image from page 9 of Andrew Ng’s mlyearning.org
*
10
Workflow of deploying a model using healthcare.ai?
• Combine features with the two algorithms and assess model performance, iteratively
• Deploy the model with the features and algorithms that worked best
11
Poll question – I/O
What type of data connections would you primarily use for R? 405 respondents
• CSV – 26%• SQL Server – 51%• MySQL – 9%• Postgres – 3%• Oracle – 11%
12
I/O for healthcare.ai
• Databases via SQL Server
• CSV, TXT, etc
13
healthcare.ai use cases
• 30-day readmissions
• Hospital acquired infections like CLABSI
• No-shows, propensity to pay, census, etc
14
Examples!
15
Roadmap• Submit to CRAN
• Submit to PyPI
• Switch to MySQL as default database
• Add deep learning into python package
16
Want to contribute?
• Clone our repos!
17
Poll question – What’s impeding you?
What is impeding you from using healthcare.ai? 277 respondents
• Loading data into R – 8%• Installing the package – 15%• Don’t know how to integrate into database infrastructure – 31%• Adoption – clinical team isn’t interested – 7%• Not sure what to predict – 38%
18
Before we end…• healthcare.ai is our public offering—it’s
currently being integrated into HC’s products
• Check out the blog
• File issues on Stack Overflow with the package version number and tag ‘healthcare-ai’
19
Questions?