Please enable JavaScript.
Coggle requires JavaScript to display documents.
Set up Prediction System (Retraining model (sometimes your model does not…
Set up Prediction System
Retraining model
sometimes your model does not perform how you wanted it to
take steps to improve predictability
when rerunning models you no longer need the holdout set
use 100% of data
Choose deployment strategy
some deployment strategies are easier to implement than others
Drag-and-drop scoring
found in the predict screen
has a size limit
you can add up to 5 features
Application Programming interface (API)
easy when you know R or python programming
allows programmer write a program to upload data and receive probabilities
allows access to reason codes used for low-volume predictions
DataRobot Prime Scoring
approximates your model in python and java and availability to you
not accurate but very close
once this approach is used you must maintain the code
prime job creates rules based on the model and suggests lowest log loss score
code is available to download
Batch Code
uses API to upload and score large files
code needed
In-place scoring
exports model into Apache Spark