willirath / ABCDE-demo

MIT License

1 stars 1 forks source link

Update notebook with observations #9

Closed ucyo closed 3 years ago

ucyo commented 3 years ago

Results

There is no major difference between Scenario A and B. Why? Possible reasons:
- Randomized data is difficult to be compressed.
- Content fill_value is very important since these can be chosen related/unrelated to the data. (e.g. Range between -2,2 and fillvalue of 50)
- zfp does use transformation-based compression and not prediction-based compression.
- 1-dim vs n-dim data can show different results.

Todo

[ ] Test with n-dim climate/ocean model data output.
[ ] Test with prediction-based compression model.

Questions

How can we relate this to AI/ML-based methods?

ucyo commented 3 years ago

Lets talk about this at our call tomorrow