brews / baysparpy Goto Github PK

Refactor Prediction.location to Prediction.latlon.

Makes things more explicit.

Consider setting up internal package version system, maybe something like the way xarray handles version numbers in setup.py and ./docs/?

Using nens in prediction functions does not evenly extract model parameter draws.

Right now, if a user wants to use a smaller ensemble of model parameter draws, we just pop the draws straight off the front. We should follow the original MATLAB code and extract from an even spread of positions (i.e. using np.linspace()).

baysparpy is huge because we've 1) all the model parameter draws for the MCMC and 2) the observation database.

At the moment the package size is almost 500 mb.

To host a package on PYPI, it needs to be 60 mb. We might be able to get away with ~ 100 mb, but we'd need to require the extra space on the repository.

On https://baysparpy.readthedocs.io/en/stable/overview.html, the subt depth is incorrectly labeled for 0 - 100 m. It should be 0 - 200 m.

This needs to be fixed in Docs.

Assertion catches in functions need to raise proper exceptions. Examples of this are at the head of the predict_*() function. If these assertions are tripped they give a very user-unfriendly error. It would be nice to give a proper error with helpful description.

Right now, baysparpy depends on cartopy. Cartopy is needed to plot the gridpoints (analogmap()) output from an analog prediction (predict_seatemp_analog()). Other components in the package do not need cartopy, just this one plotting function.

Cartopy can be a pain to install, especially for new users.

Might want to make cartopy optional for installs, note this in the documentation and print out an error message on import saying that cartopy needs to be installed... something like this.

For predict_tex() and predict_seatemp(), the docstrings say that we're returning a dictionary insead of a Prediction object. This needs to be fixed.

I think model parameter draws for tau2 has some extra draws because burnin was not removed. We need to be sure that all draws are equal (~20,000 draws) and the correct burnin is removed.

Need conda install instructions. Use the 'sbmalev' channel.

Consider subclassing Prediction into TexPrediction, SeatempPrediction, AnalogPrediction or similar.

I feel like too many of our attributes are None for some of these prediction types.

README needs to be put as long_description in setup.py.