def build_suggest_ds()

in bayesmark/experiment.py [0:0]

• 12 lines of code
• 6 McCabe index (conditional complexity)

def build_suggest_ds(suggest_log):
    """Convert :class:`numpy:numpy.ndarray` with function evaluation inputs to :class:`xarray:xarray.Dataset`.

    This function is a data cleanup routine after running an experiment, before serializing the data to end the study.

    Parameters
    ----------
    suggest_log : list(list(dict(str, object)))
        Log of the suggestions. It has shape `(n_call, n_suggest)`.

    Returns
    -------
    suggest_ds : :class:`xarray:xarray.Dataset`
        :class:`xarray:xarray.Dataset` containing one variable for each input with the objective function evaluations.
        It has dimensions ``(ITER, SUGGEST)``.
    """
    n_call, n_suggest = np.shape(suggest_log)
    assert n_call * n_suggest > 0

    # Setup the dims
    ds_vars = sorted(suggest_log[0][0].keys())
    coords = OrderedDict([(ITER, range(n_call)), (SUGGEST, range(n_suggest))])

    # There is prob a way to vectorize this more but good enough for now. Using np.full to infer dtype from 1st element
    data = OrderedDict([(kk, ((ITER, SUGGEST), np.full((n_call, n_suggest), suggest_log[0][0][kk]))) for kk in ds_vars])
    for ii in range(n_call):
        for jj in range(n_suggest):
            for kk in ds_vars:
                data[kk][1][ii, jj] = suggest_log[ii][jj][kk]

    suggest_ds = xr.Dataset(data, coords=coords)
    return suggest_ds