tommylees112
commented
5 years ago
What exactly is the EventDetector doing? The notebook says that it identifies any pixel that exceeds a threshold. How do you define the thresholds?
The EventDetector allows a user to flexibly identify consecutive periods of above/below threshold timesteps. The thresholds can either be supplied or inferred from the data (Q10, Q90, mean +- 1std etc.). It’s purpose in the pipeline is for prior and posterior analysis of severity and it is relatively self-contained. If necessary we can move it into a experimental location but its role is to help interpret the input/output variables.
Could you please elaborate more on the interpretation of your results?
Of course! When you say interpretation are you referring mostly to the feature importances and the times/places where the model performs well or badly?
We are currently trying to push for a highly performant model given the data that we already have. From this we will be able to demonstrate the interpretation of our results more fully. We are also working on a class that calculates the performance of the model by different groupings (e.g. landcover classes or administrative regions).
What exactly is the EventDetector doing? The notebook says that it identifies any pixel that exceeds a threshold. How do you define the thresholds?
Could you please elaborate more on the interpretation of your results?