Please enable JavaScript.
Coggle requires JavaScript to display documents.
Set up a Prediction system (Choose deployment strategy (In-place scoring…
Set up a Prediction system
Help from corporate IT department most likely necessary
Retraining model
Model may perform worse than expected when put into practice
One reason for lower success rate: environment being model changes subtly
Can improve predictive ability by
Finding top model, clicking orange plus sign
Run model with full sample (unlock holdout)
Choose deployment strategy
Drag and drop scoring
Accessed through predict screen
Upload file (test data) then click compute predictions
Can add up to 5 features
Recommend to remove ID column
Data robot prime scoring
Creates an approximation of selected model
Available as code in python and java
Data robot cannot promise its as accurate as the original model but its comparable
Batch scoring
uses API to upload/score multiple large files in parallel
Code found at:
https://github.com/datarobot/batch-scoring
.
In-place scoring
exports model as an executable file to use in Apache Spark
Spark is a fast data processing environment
Ideal for companies already using Spark cluster
API (Application Programming Interface)
created on DataRobot which allows a developer to write a program