AutoAI feature comparison
Configuration options for AutoAI experiments differ based on the type of experiment you build. This table notes the differences between options for various types of AutoAI experiments.
Experiment type
An AutoAI experiment can be one of these types:
- Binary classification, multiclass classification, or regression model trained by using a single data source.
- Binary classification, multiclass classification, or regression model that is built from joining multiple data sources to train the experiment.
- Univariate or multivariate time series forecast experiment.
Feature support and configuration options by experiment type
Feature or option | Single data source | Joined data | Time series |
---|---|---|---|
Max number of data files | 1 | 4 | 1 |
Testing data file | ✓ | ||
Training data limits | 1 GB | 4 GB per file, 20 GB total | Max of 10 columns, 300,000 rows |
Define holdout size | ✓ (percent in the range 85 - 95%) | ✓ (percent 85 - 95%) | ✓ (as number of rows between 1 and dataLength/2) |
Max number of classes | 200 | 200 | |
Drop Duplicate Rows | ✓ | ||
Subsample data | ✓ | ✓ | |
Exclude columns from training | ✓ | ✓ | ✓ |
Text transformer | ✓ | ||
CSV as input format | ✓ | ✓ | ✓ |
Fairness evaluation | ✓ | ||
Stratified sampling limit | ✓ | ||
Sliding window | ✓ | ||
Cutoff Timestamp | ✓ | ||
Deduplication (remove duplicated features) | ✓ | ||
Inconsistency (remove features with inconsistent distribution) | ✓ | ||
Filter (remove low correlations) | ✓ | ||
Remove missing target rows | ✓ | ||
Remove category features with unique (only) values | ✓ | ||
Remove features with constant values | ✓ | ||
Supported timestamp formats | ✓ | ✓ | |
Forecast window | ✓ | ||
Lookback window | ✓ | ||
Number of backtests | ✓ | ||
Number of prediction columns | 1 | 1 | up to 10 |
Generated experiment notebook | ✓ | ✓ | |
Online deployment | ✓ | ✓ | |
Batch deployment | ✓ | ✓ | ✓ |
Learn more
Parent topic: AutoAI