Please enable JavaScript.
Coggle requires JavaScript to display documents.
Ch. 21: Set Up Prediction System (Choose Deployment Strategy (Drag and…
Ch. 21: Set Up Prediction System
Retraining Model
model's success rate can be lower than expected
environment changing subtly over time
was validated w/ data collected during same time period
can re-run models with 100% of data set
Choose Deployment Strategy
Drag and Drop Scoring
slow approach
does not allow files larger than 750MB
accessed through Predict screen of selected model
API Scoring
Application Program Interface (API) created on Datarobot server
useful for those who can code in python or R
allows for access to reason codes
important for low-volume predictions
DataRobot Prime Scoring
creates approximation of selected model
available as python and java code
con: in-house responsibility to integrate and maintain model code
pro: vendor independence, allows use of organization's own servers
Batch Scoring
uses DataRobot API to upload and score multiple files in parallel
In-Place Scoring
exports model for use in Apache Spark environment