Learn how to upsample your model to increase the sampling factor used for the Upper and Lower Percentiles of your model
What is upsampling?
Upsampling within the platform allows you to increase the number of occurrences that your model encounters of your outlier locations or of a sample of a more unique store type . This should as a result provide the model with further learning on these types of locations and therefore can cause the model's ability to provide more accurate future forecasts for these type of scenarios.
Increasing the Sampling Factor increases the number of upsampled occurrences on each Upper Percentile and Lower Percentile end with a cap of 50 for the largest number of upsamples.
The goal of upsampling is to find the balance of which Upper Percentile value and which Lower Percentile value allows you to to increase the number of occurrences of outliers that the model has to learn from without inflating the number of mid range performers.
How do I upsample my sites?
To upsample your sites you will need to set the following three criteria:
- Lower Percentile: Lowest performing percentile with possible settings from 3% to 20%. Generally the higher the percentage the more locations fall within that percentile.
- Upper Percentile: Highest performing percentile with possible settings from 80% to 98%. Generally the higher the percentage the fewer location fall within that percentile.
- Sampling Factor: Number of occurrences that you would like to upsample each end by. For example a sampling factor of 3 would provide the model with 3 additional sample for each site that falls within your Upper Percentile and Lower Percentile groups.