arviz 0.19.0__py3-none-any.whl → 0.21.0__py3-none-any.whl

arviz/stats/stats.py

@@ -27,6 +27,7 @@ from ..utils import Numba, _numba_var, _var_names, get_coords
 from .density_utils import get_bins as _get_bins
 from .density_utils import histogram as _histogram
 from .density_utils import kde as _kde
+from .density_utils import _kde_linear
 from .diagnostics import _mc_error, _multichain_statistics, ess
 from .stats_utils import ELPDData, _circular_standard_deviation, smooth_data
 from .stats_utils import get_log_likelihood as _get_log_likelihood
@@ -41,6 +42,7 @@ from ..labels import BaseLabeller
 
 __all__ = [
     "apply_test_function",
+    "bayes_factor",
     "compare",
     "hdi",
     "loo",
@@ -711,16 +713,19 @@ def loo(data, pointwise=None, var_name=None, reff=None, scale=None):
     Returns
     -------
     ELPDData object (inherits from :class:`pandas.Series`) with the following row/attributes:
-    elpd: approximated expected log pointwise predictive density (elpd)
+    elpd_loo: approximated expected log pointwise predictive density (elpd)
     se: standard error of the elpd
     p_loo: effective number of parameters
-    shape_warn: bool
-        True if the estimated shape parameter of Pareto distribution is greater than a thresold
-        value  for one or more samples. For a sample size S, the thresold is compute as
-        min(1 - 1/log10(S), 0.7)
-    loo_i: array of pointwise predictive accuracy, only if pointwise True
+    n_samples: number of samples
+    n_data_points: number of data points
+    warning: bool
+        True if the estimated shape parameter of Pareto distribution is greater than
+        ``good_k``.
+    loo_i: :class:`~xarray.DataArray` with the pointwise predictive accuracy,
+            only if pointwise=True
     pareto_k: array of Pareto shape values, only if pointwise True
     scale: scale of the elpd
+    good_k: For a sample size S, the thresold is compute as min(1 - 1/log10(S), 0.7)
 
         The returned object has a custom print method that overrides pd.Series method.
 
@@ -914,6 +919,7 @@ def psislw(log_weights, reff=1.0):
            ...: az.psislw(-log_likelihood, reff=0.8)
 
     """
+    log_weights = deepcopy(log_weights)
     if hasattr(log_weights, "__sample__"):
         n_samples = len(log_weights.__sample__)
         shape = [
@@ -1580,7 +1586,9 @@ def waic(data, pointwise=None, var_name=None, scale=None, dask_kwargs=None):
     elpd_waic: approximated expected log pointwise predictive density (elpd)
     se: standard error of the elpd
     p_waic: effective number parameters
-    var_warn: bool
+    n_samples: number of samples
+    n_data_points: number of data points
+    warning: bool
         True if posterior variance of the log predictive densities exceeds 0.4
     waic_i: :class:`~xarray.DataArray` with the pointwise predictive accuracy,
             only if pointwise=True
@@ -2331,3 +2339,87 @@ def _cjs_dist(draws, weights):
     bound = cdf_p_int + cdf_q_int
 
     return np.sqrt((cjs_pq + cjs_qp) / bound)
+
+
+def bayes_factor(idata, var_name, ref_val=0, prior=None, return_ref_vals=False):
+    r"""Approximated Bayes Factor for comparing hypothesis of two nested models.
+
+    The Bayes factor is estimated by comparing a model (H1) against a model in which the
+    parameter of interest has been restricted to be a point-null (H0). This computation
+    assumes the models are nested and thus H0 is a special case of H1.
+
+    Notes
+    -----
+    The bayes Factor is approximated as the Savage-Dickey density ratio
+    algorithm presented in [1]_.
+
+    Parameters
+    ----------
+    idata : InferenceData
+        Any object that can be converted to an :class:`arviz.InferenceData` object
+        Refer to documentation of :func:`arviz.convert_to_dataset` for details.
+    var_name : str, optional
+        Name of variable we want to test.
+    ref_val : int, default 0
+        Point-null for Bayes factor estimation.
+    prior : numpy.array, optional
+        In case we want to use different prior, for example for sensitivity analysis.
+    return_ref_vals : bool, optional
+        Whether to return the values of the prior and posterior at the reference value.
+        Used by :func:`arviz.plot_bf` to display the distribution comparison.
+
+
+    Returns
+    -------
+    dict : A dictionary with BF10 (Bayes Factor 10 (H1/H0 ratio), and BF01 (H0/H1 ratio).
+
+    References
+    ----------
+    .. [1] Heck, D., 2019. A caveat on the Savage-Dickey density ratio:
+       The case of computing Bayes factors for regression parameters.
+
+    Examples
+    --------
+    Moderate evidence indicating that the parameter "a" is different from zero.
+
+    .. ipython::
+
+        In [1]: import numpy as np
+           ...: import arviz as az
+           ...: idata = az.from_dict(posterior={"a":np.random.normal(1, 0.5, 5000)},
+           ...: prior={"a":np.random.normal(0, 1, 5000)})
+           ...: az.bayes_factor(idata, var_name="a", ref_val=0)
+
+    """
+
+    posterior = extract(idata, var_names=var_name).values
+
+    if ref_val > posterior.max() or ref_val < posterior.min():
+        _log.warning(
+            "The reference value is outside of the posterior. "
+            "This translate into infinite support for H1, which is most likely an overstatement."
+        )
+
+    if posterior.ndim > 1:
+        _log.warning("Posterior distribution has {posterior.ndim} dimensions")
+
+    if prior is None:
+        prior = extract(idata, var_names=var_name, group="prior").values
+
+    if posterior.dtype.kind == "f":
+        posterior_grid, posterior_pdf, *_ = _kde_linear(posterior)
+        prior_grid, prior_pdf, *_ = _kde_linear(prior)
+        posterior_at_ref_val = np.interp(ref_val, posterior_grid, posterior_pdf)
+        prior_at_ref_val = np.interp(ref_val, prior_grid, prior_pdf)
+
+    elif posterior.dtype.kind == "i":
+        posterior_at_ref_val = (posterior == ref_val).mean()
+        prior_at_ref_val = (prior == ref_val).mean()
+
+    bf_10 = prior_at_ref_val / posterior_at_ref_val
+    bf = {"BF10": bf_10, "BF01": 1 / bf_10}
+
+    if return_ref_vals:
+        return (bf, {"prior": prior_at_ref_val, "posterior": posterior_at_ref_val})
+    else:
+        return bf

arviz/tests/base_tests/test_data.py

@@ -3,6 +3,7 @@
 
 import importlib
 import os
+import warnings
 from collections import namedtuple
 from copy import deepcopy
 from html import escape
@@ -32,7 +33,13 @@ from ... import (
     extract,
 )
 
-from ...data.base import dict_to_dataset, generate_dims_coords, infer_stan_dtypes, make_attrs
+from ...data.base import (
+    dict_to_dataset,
+    generate_dims_coords,
+    infer_stan_dtypes,
+    make_attrs,
+    numpy_to_data_array,
+)
 from ...data.datasets import LOCAL_DATASETS, REMOTE_DATASETS, RemoteFileMetadata
 from ..helpers import (  # pylint: disable=unused-import
     chains,
@@ -44,6 +51,10 @@ from ..helpers import (  # pylint: disable=unused-import
     models,
 )
 
+# Check if dm-tree is installed
+dm_tree_installed = importlib.util.find_spec("tree") is not None  # pylint: disable=invalid-name
+skip_tests = (not dm_tree_installed) and ("ARVIZ_REQUIRE_ALL_DEPS" not in os.environ)
+
 
 @pytest.fixture(autouse=True)
 def no_remote_data(monkeypatch, tmpdir):
@@ -227,6 +238,17 @@ def test_dims_coords_skip_event_dims(shape):
     assert "z" not in coords
 
 
+@pytest.mark.parametrize("dims", [None, ["chain", "draw"], ["chain", "draw", None]])
+def test_numpy_to_data_array_with_dims(dims):
+    da = numpy_to_data_array(
+        np.empty((4, 500, 7)),
+        var_name="a",
+        dims=dims,
+        default_dims=["chain", "draw"],
+    )
+    assert list(da.dims) == ["chain", "draw", "a_dim_0"]
+
+
 def test_make_attrs():
     extra_attrs = {"key": "Value"}
     attrs = make_attrs(attrs=extra_attrs)
@@ -895,6 +917,11 @@ class TestInferenceData:  # pylint: disable=too-many-public-methods
         assert escape(repr(idata)) in html
         xr.set_options(display_style=display_style)
 
+    def test_setitem(self, data_random):
+        data_random["new_group"] = data_random.posterior
+        assert "new_group" in data_random.groups()
+        assert hasattr(data_random, "new_group")
+
     def test_add_groups(self, data_random):
         data = np.random.normal(size=(4, 500, 8))
         idata = data_random
@@ -912,7 +939,7 @@ class TestInferenceData:  # pylint: disable=too-many-public-methods
         data = np.random.normal(size=(4, 500, 8))
         idata = data_random
         with pytest.warns(UserWarning, match="The group.+not defined in the InferenceData scheme"):
-            idata.add_groups({"new_group": idata.posterior})
+            idata.add_groups({"new_group": idata.posterior}, warn_on_custom_groups=True)
         with pytest.warns(UserWarning, match="the default dims.+will be added automatically"):
             idata.add_groups(constant_data={"a": data[..., 0], "b": data})
         assert idata.new_group.equals(idata.posterior)
@@ -953,8 +980,8 @@ class TestInferenceData:  # pylint: disable=too-many-public-methods
         with pytest.raises(ValueError, match="join must be either"):
             idata.extend(idata2, join="outer")
         idata2.add_groups(new_group=idata2.prior)
-        with pytest.warns(UserWarning):
-            idata.extend(idata2)
+        with pytest.warns(UserWarning, match="new_group"):
+            idata.extend(idata2, warn_on_custom_groups=True)
 
 
 class TestNumpyToDataArray:
@@ -1076,6 +1103,7 @@ def test_dict_to_dataset():
     assert set(dataset.b.coords) == {"chain", "draw", "c"}
 
 
+@pytest.mark.skipif(skip_tests, reason="test requires dm-tree which is not installed")
 def test_nested_dict_to_dataset():
     datadict = {
         "top": {"a": np.random.randn(100), "b": np.random.randn(1, 100, 10)},
@@ -1146,11 +1174,17 @@ def test_bad_inference_data():
         InferenceData(posterior=[1, 2, 3])
 
 
-def test_inference_data_other_groups():
+@pytest.mark.parametrize("warn", [True, False])
+def test_inference_data_other_groups(warn):
     datadict = {"a": np.random.randn(100), "b": np.random.randn(1, 100, 10)}
     dataset = convert_to_dataset(datadict, coords={"c": np.arange(10)}, dims={"b": ["c"]})
-    with pytest.warns(UserWarning, match="not.+in.+InferenceData scheme"):
-        idata = InferenceData(other_group=dataset)
+    if warn:
+        with pytest.warns(UserWarning, match="not.+in.+InferenceData scheme"):
+            idata = InferenceData(other_group=dataset, warn_on_custom_groups=True)
+    else:
+        with warnings.catch_warnings():
+            warnings.simplefilter("error")
+            idata = InferenceData(other_group=dataset, warn_on_custom_groups=False)
     fails = check_multiple_attrs({"other_group": ["a", "b"]}, idata)
     assert not fails
 

arviz/tests/base_tests/test_diagnostics.py

@@ -120,10 +120,11 @@ class TestDiagnostics:
         ```
         Reference file:
 
-        Created: 2020-08-31
-        System: Ubuntu 18.04.5 LTS
-        R version 4.0.2 (2020-06-22)
-        posterior 0.1.2
+        Created: 2024-12-20
+        System: Ubuntu 24.04.1 LTS
+        R version 4.4.2 (2024-10-31)
+        posterior version from https://github.com/stan-dev/posterior/pull/388
+            (after release 1.6.0 but before the fixes in the PR were released).
         """
         # download input files
         here = os.path.dirname(os.path.abspath(__file__))

arviz/tests/base_tests/test_plots_matplotlib.py

@@ -2,21 +2,22 @@
 
 # pylint: disable=redefined-outer-name,too-many-lines
 import os
+import re
 from copy import deepcopy
 
 import matplotlib.pyplot as plt
 import numpy as np
 import pytest
+import xarray as xr
 from matplotlib import animation
 from pandas import DataFrame
 from scipy.stats import gaussian_kde, norm
-import xarray as xr
 
 from ...data import from_dict, load_arviz_data
 from ...plots import (
     plot_autocorr,
-    plot_bpv,
     plot_bf,
+    plot_bpv,
     plot_compare,
     plot_density,
     plot_dist,
@@ -43,20 +44,20 @@ from ...plots import (
     plot_ts,
     plot_violin,
 )
+from ...plots.dotplot import wilkinson_algorithm
+from ...plots.plot_utils import plot_point_interval
 from ...rcparams import rc_context, rcParams
 from ...stats import compare, hdi, loo, waic
 from ...stats.density_utils import kde as _kde
-from ...utils import _cov, BehaviourChangeWarning
-from ...plots.plot_utils import plot_point_interval
-from ...plots.dotplot import wilkinson_algorithm
+from ...utils import BehaviourChangeWarning, _cov
 from ..helpers import (  # pylint: disable=unused-import
+    RandomVariableTestClass,
     create_model,
     create_multidimensional_model,
     does_not_warn,
     eight_schools_params,
     models,
     multidim_models,
-    RandomVariableTestClass,
 )
 
 rcParams["data.load"] = "eager"
@@ -1236,6 +1237,23 @@ def test_plot_hdi_dataset_error(models):
         plot_hdi(np.arange(8), hdi_data=hdi_data)
 
 
+def test_plot_hdi_string_error():
+    """Check x as type string raises an error."""
+    x_data = ["a", "b", "c", "d"]
+    y_data = np.random.normal(0, 5, (1, 200, len(x_data)))
+    hdi_data = hdi(y_data)
+    with pytest.raises(
+        NotImplementedError,
+        match=re.escape(
+            (
+                "The `arviz.plot_hdi()` function does not support categorical data. "
+                "Consider using `arviz.plot_forest()`."
+            )
+        ),
+    ):
+        plot_hdi(x=x_data, y=y_data, hdi_data=hdi_data)
+
+
 def test_plot_hdi_datetime_error():
     """Check x as datetime raises an error."""
     x_data = np.arange(start="2022-01-01", stop="2022-03-01", dtype=np.datetime64)
@@ -1285,10 +1303,11 @@ def test_plot_ecdf_eval_points():
     assert axes is not None
 
 
-@pytest.mark.parametrize("confidence_bands", [True, "pointwise", "simulated"])
-def test_plot_ecdf_confidence_bands(confidence_bands):
+@pytest.mark.parametrize("confidence_bands", [True, "pointwise", "optimized", "simulated"])
+@pytest.mark.parametrize("ndraws", [100, 10_000])
+def test_plot_ecdf_confidence_bands(confidence_bands, ndraws):
     """Check that all confidence_bands values correctly accepted"""
-    data = np.random.randn(4, 1000)
+    data = np.random.randn(4, ndraws // 4)
     axes = plot_ecdf(data, confidence_bands=confidence_bands, cdf=norm(0, 1).cdf)
     assert axes is not None
 
@@ -1326,6 +1345,8 @@ def test_plot_ecdf_error():
     # contradictory confidence band types
     with pytest.raises(ValueError):
         plot_ecdf(data, cdf=dist.cdf, confidence_bands="simulated", pointwise=True)
+    with pytest.raises(ValueError):
+        plot_ecdf(data, cdf=dist.cdf, confidence_bands="optimized", pointwise=True)
     plot_ecdf(data, cdf=dist.cdf, confidence_bands=True, pointwise=True)
     plot_ecdf(data, cdf=dist.cdf, confidence_bands="pointwise")
 
@@ -2079,7 +2100,5 @@ def test_plot_bf():
     idata = from_dict(
         posterior={"a": np.random.normal(1, 0.5, 5000)}, prior={"a": np.random.normal(0, 1, 5000)}
     )
-    bf_dict0, _ = plot_bf(idata, var_name="a", ref_val=0)
-    bf_dict1, _ = plot_bf(idata, prior=np.random.normal(0, 10, 5000), var_name="a", ref_val=0)
-    assert bf_dict0["BF10"] > bf_dict0["BF01"]
-    assert bf_dict1["BF10"] < bf_dict1["BF01"]
+    _, bf_plot = plot_bf(idata, var_name="a", ref_val=0)
+    assert bf_plot is not None

arviz/tests/base_tests/test_stats.py

@@ -18,6 +18,7 @@ from ...data import concat, convert_to_inference_data, from_dict, load_arviz_dat
 from ...rcparams import rcParams
 from ...stats import (
     apply_test_function,
+    bayes_factor,
     compare,
     ess,
     hdi,
@@ -871,3 +872,13 @@ def test_priorsens_coords(psens_data):
     assert "mu" in result
     assert "theta" in result
     assert "school" in result.theta_t.dims
+
+
+def test_bayes_factor():
+    idata = from_dict(
+        posterior={"a": np.random.normal(1, 0.5, 5000)}, prior={"a": np.random.normal(0, 1, 5000)}
+    )
+    bf_dict0 = bayes_factor(idata, var_name="a", ref_val=0)
+    bf_dict1 = bayes_factor(idata, prior=np.random.normal(0, 10, 5000), var_name="a", ref_val=0)
+    assert bf_dict0["BF10"] > bf_dict0["BF01"]
+    assert bf_dict1["BF10"] < bf_dict1["BF01"]

arviz/tests/base_tests/test_stats_ecdf_utils.py

@@ -10,6 +10,13 @@ from ...stats.ecdf_utils import (
     _get_pointwise_confidence_band,
 )
 
+try:
+    import numba  # pylint: disable=unused-import
+
+    numba_options = [True, False]
+except ImportError:
+    numba_options = [False]
+
 
 def test_compute_ecdf():
     """Test compute_ecdf function."""
@@ -109,9 +116,15 @@ def test_get_pointwise_confidence_band(dist, prob, ndraws, num_trials=1_000, see
     ids=["continuous", "continuous default rvs", "discrete"],
 )
 @pytest.mark.parametrize("ndraws", [10_000])
-@pytest.mark.parametrize("method", ["pointwise", "simulated"])
-def test_ecdf_confidence_band(dist, rvs, prob, ndraws, method, num_trials=1_000, seed=57):
+@pytest.mark.parametrize("method", ["pointwise", "optimized", "simulated"])
+@pytest.mark.parametrize("use_numba", numba_options)
+def test_ecdf_confidence_band(
+    dist, rvs, prob, ndraws, method, use_numba, num_trials=1_000, seed=57
+):
     """Test test_ecdf_confidence_band."""
+    if use_numba and method != "optimized":
+        pytest.skip("Numba only used in optimized method")
+
     eval_points = np.linspace(*dist.interval(0.99), 10)
     cdf_at_eval_points = dist.cdf(eval_points)
     random_state = np.random.default_rng(seed)

arviz/utils.py

@@ -330,6 +330,7 @@ class Numba:
     """A class to toggle numba states."""
 
     numba_flag = numba_check()
+    """bool: Indicates whether Numba optimizations are enabled. Defaults to False."""
 
     @classmethod
     def disable_numba(cls):
@@ -732,7 +733,10 @@ class Dask:
     """
 
     dask_flag = False
+    """bool: Enables Dask parallelization when set to True. Defaults to False."""
     dask_kwargs = None
+    """dict: Additional keyword arguments for Dask configuration.
+    Defaults to an empty dictionary."""
 
     @classmethod
     def enable_dask(cls, dask_kwargs=None):

{arviz-0.19.0.dist-info → arviz-0.21.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: arviz
-Version: 0.19.0
+Version: 0.21.0
 Summary: Exploratory analysis of Bayesian models
 Home-page: http://github.com/arviz-devs/arviz
 Author: ArviZ Developers
@@ -21,30 +21,30 @@ Classifier: Topic :: Scientific/Engineering :: Mathematics
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: setuptools >=60.0.0
-Requires-Dist: matplotlib >=3.5
-Requires-Dist: numpy >=1.23.0
-Requires-Dist: scipy >=1.9.0
+Requires-Dist: setuptools>=60.0.0
+Requires-Dist: matplotlib>=3.5
+Requires-Dist: numpy>=1.23.0
+Requires-Dist: scipy>=1.9.0
 Requires-Dist: packaging
-Requires-Dist: pandas >=1.5.0
-Requires-Dist: dm-tree >=0.1.8
-Requires-Dist: xarray >=2022.6.0
-Requires-Dist: h5netcdf >=1.0.2
-Requires-Dist: typing-extensions >=4.1.0
-Requires-Dist: xarray-einstats >=0.3
+Requires-Dist: pandas>=1.5.0
+Requires-Dist: xarray>=2022.6.0
+Requires-Dist: h5netcdf>=1.0.2
+Requires-Dist: typing-extensions>=4.1.0
+Requires-Dist: xarray-einstats>=0.3
 Provides-Extra: all
-Requires-Dist: numba ; extra == 'all'
-Requires-Dist: netcdf4 ; extra == 'all'
-Requires-Dist: bokeh >=3 ; extra == 'all'
-Requires-Dist: contourpy ; extra == 'all'
-Requires-Dist: ujson ; extra == 'all'
-Requires-Dist: dask[distributed] ; extra == 'all'
-Requires-Dist: zarr <3,>=2.5.0 ; extra == 'all'
-Requires-Dist: xarray-datatree ; extra == 'all'
+Requires-Dist: numba; extra == "all"
+Requires-Dist: netcdf4; extra == "all"
+Requires-Dist: bokeh>=3; extra == "all"
+Requires-Dist: contourpy; extra == "all"
+Requires-Dist: ujson; extra == "all"
+Requires-Dist: dask[distributed]; extra == "all"
+Requires-Dist: zarr<3,>=2.5.0; extra == "all"
+Requires-Dist: xarray-datatree; extra == "all"
+Requires-Dist: dm-tree>=0.1.8; extra == "all"
 Provides-Extra: preview
-Requires-Dist: arviz-base[h5netcdf] ; extra == 'preview'
-Requires-Dist: arviz-stats[xarray] ; extra == 'preview'
-Requires-Dist: arviz-plots ; extra == 'preview'
+Requires-Dist: arviz-base[h5netcdf]; extra == "preview"
+Requires-Dist: arviz-stats[xarray]; extra == "preview"
+Requires-Dist: arviz-plots; extra == "preview"
 
 <img src="https://raw.githubusercontent.com/arviz-devs/arviz-project/main/arviz_logos/ArviZ.png#gh-light-mode-only" width=200></img>
 <img src="https://raw.githubusercontent.com/arviz-devs/arviz-project/main/arviz_logos/ArviZ_white.png#gh-dark-mode-only" width=200></img>