Please enable JavaScript.
Coggle requires JavaScript to display documents.
Understanding the Process (Learning Curves and Speed (leaderboard tab (log…
Understanding the Process
Learning Curves and Speed
More data? Improve?
additional features/additional cases
leaderboard tab
learning curves
log loss for accuracy
graph in an elbow shape
click to activate, keep
Apply numbers to reality
Models improve as they get more data
calculate cost of if data is available
humans are hard to distinctly calculate
Accuracy Trade Offs
Speed versus accuracy
How rapidly model will evaluate
Efficient frontier
speed and accuracy - correlated?
Model should be efficient
new data, new prediction
top 10 based on validation score
slowest model, calculate
most accurate?
speed as a criteria
necessary?
if irrelevant, skip this
Blueprints
Every model utilizes preprocessing steps
secret sauce for data robot
Imputation
Regularized Logistic Models
missing values imputed
data robot model docs
standardization
some features struggle with different standard deviations
scaled - mean value = 0
one hot encoding
Yes/No
minimum records
minsupport
not mapped
card-min/card-max
algorithm box
hyperparameter content
advanced