Please enable JavaScript.
Coggle requires JavaScript to display documents.
CH 21-23 (Model Documentation (Critical to justify and note steps for…
CH 21-23
Model Documentation
Critical to justify and note steps for future purpose/use
where did data come from, how was it processed, what parameters were used, use in the business?
assume project will be revisited within a year
Set up prediction system
model is given current data to make a prediction
holdout logloss < CV score indicates performance under expectations
environment being modeled changes slightly over time
models can be created with 100% of data when retraining
deployment strategies:
drag and drop- (easiest) compute predictions
API scoring - update code as data comes
DataRobot Prime, Batch Scoring, In-Place with Spark
Create a model monitoring and maintenance plan
DataRobot prediction fails instead of making best of data
rerun model as soon as sufficient new data is available
evaluate training data against new data
retrain if model can distinguish the different data
use Mathews Correlation Coefficient