Add autocorrelation plot #153
Conversation
src/arviz_plots/plot_collection.py
Outdated
| # def get_target(self, var_name, selection): | ||
| # """Get the target that corresponds to the given variable and selection.""" | ||
| # return subset_ds(self.get_viz(var_name), "plot", selection) | ||
|
|
There was a problem hiding this comment.
Don't change this, focus on the plot
| plot_collection : PlotCollection | ||
| The plot collection containing the autocorrelation plots. | ||
| """ | ||
| dt = convert_to_dataset(dt, group="posterior") |
There was a problem hiding this comment.
| dt = convert_to_dataset(dt, group="posterior") |
|
|
||
| # Calculate autocorrelation from arviz_stats autocorr computation | ||
| core_base = _CoreBase() | ||
| acf_data = core_base.autocorr(distribution_array, axis=-1)[..., :max_lag] |
There was a problem hiding this comment.
If autocorr is not an accessor it should be, then you will be able to pass ds/da.
There was a problem hiding this comment.
sir, you mentioned making autocorr work with xarray objects. Should I :
-Raise an issue in arviz_stats to modify autocorr for xarray support,
or
-Create a wrapper/accessor in autocorrplot.py locally?
| n_plots, plots_per_var = process_facet_dims(pc_data, pc_kwargs["cols"]) | ||
|
|
||
| # Set up figure size | ||
| figsize = pc_kwargs.get("plot_grid_kws", {}).get("figsize", None) |
There was a problem hiding this comment.
You can copy all the "if plot_collection is None" logic from other plots like plotdist
|
@aloctavodia sir, you mentioned making autocorr work with xarray objects. Should I : |
|
Arviz-stats already have accessors for a few functions like hdi, ess, etc. So that's the place to implement them. |
|
@suhaani-agarwal are you still working on this? |
yes, I am still working on this. I am stuck in some dimension related properties of the autocorr computation. |
aloctavodia
changed the title
|
Hello , @aloctavodia this is some progress on the code....I have added the computation using the autocorr accessor, still a lot of work is remaining. I have some doubts...
|
|
It is incorrect to do the looping manually. See for example https://github.com/arviz-devs/arviz-plots/blob/main/src/arviz_plots/plots/ecdfplot.py, |
This is in reference to Issue #2
Current State:
acf_datais being computed correctly from autocorr function fromarviz-stats, but there are warnings about missing selections inviz[autocorr]-Warning: Selection (('chain', np.int64(0)), ('draw', np.int64(0)), ('lag', np.int64(0))) not found in viz[autocorr].This indicates a mismatch in the expected dimensions or coordinates in the
vizdataset.I would greatly appreciate your review and guidance on any changes needed to address errors or improve the implementation.
📚 Documentation preview 📚: https://arviz-plots--153.org.readthedocs.build/en/153/