astrofrog / fast-histogram

:zap: Fast 1D and 2D histogram functions in Python :zap:
BSD 2-Clause "Simplified" License
266 stars 28 forks source link

Tropical Rainfall Rebinning and Error Band Issues #88

Closed NataNazar closed 7 months ago

NataNazar commented 7 months ago
  1. Histogram Rebinning with Incorrect Normalization

    • Problem: When rebinning histograms, the normalization appears incorrect, leading to inaccurate PDF distributions. Currently, rebinning is disabled in the command-line interface (CLI) to avoid this issue.
    • Action Required: Investigate the rebinning logic to determine the root cause of the incorrect normalization. It involves:
    • Checking the number of counts in the histogram before and after rebinning.
  2. Error Band for Daily Variability

    • Problem: The error band calculation for daily variability uses a fitting method, which may not be optimal. The current method could lead to an inaccurate representation of variability error band.
    • Action Required: Adjust the error band calculation by:
    • Removing the fitting step to avoid potential biases.
    • Using the 95th percentile for error band calculation instead of the 75th percentile. This change aims to capture a broader range of variability, providing a more robust representation.
  3. Also:

    • Add error bars!
    • Be sure that the plot is not empty while you are saving it.
    • Proposed deletion of the diagnostics/tropical_rainfall/cli/run_cli_tropical_rainfall.sh script.
    • Implemented a warning system to alert users of potentially lengthy computation times, providing an estimated duration for better planning and expectation management.