Skip to content

Conversation

@sorenia
Copy link

@sorenia sorenia commented Feb 19, 2021

Key developments to whakaari code

  • Earthquake Filtration
  • Confidence Intervals
  • zscore transform
  • probability calibrations

bobdachef and others added 8 commits February 4, 2021 16:27
Earthquake Filter & uncertainty analysis
To prevent immediate errors with current forecasters I added back Ncl.
- Once everyone hardcodes or works around this can remove from train.
The exclude dates in the ForecastModel call is only used to calculate 'zsc_X'
- It is solely used to hide TEST data when calculatig the z score,
- otherwise the eruption that was supposed to be "hidden" would skew the z score
- ensure that if 'zsc_X' is in datastream MUST have necessary exclude_dates in ForecastModel() call
- The same exclude_dates should be used in .train()

USAGE:
fm = ForecastModel(..., data_streams=['zsc_rsam'],
   exclude_dates=[[tei,tef],], ... )
fm.train(..., exclude_dates=[[tei,tef],], ...)

NOTE: This is part of Annie Li's summer research 2020/21
By changing the __init__() remember to make changes to .hires_forecast()
 - a ForecastModel is created using positional arguments

NOTE: Consider updating .hires_forecast() with name based arguments
- sig_params in forecast, hires_forecast to represent a and b
- sigmoid() function
  - Return the calibrated probability by applying the sigmoid function to the model consensus.
- plot_forecast_calibrated for probability version of plot_forecast
  - changes to y limits and threshold
- similarly plot_hires_forecast_calibrated for plot_hires_forecast
  - important changes to compute_CI implementation
- same paradigm as forecast_dec2019()
- sig_params calculated in s_calibration branch
* allows for stacking of ensemble mean and probability
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants