More fixes thanks to tests

bambi/backend/pymc.py

@@ -371,12 +371,12 @@ def _run_laplace(self, draws, omit_offsets, include_mean):
 
         Parameters
         ----------
-        draws: int
+        draws : int
             The number of samples to draw from the posterior distribution.
-        omit_offsets: bool
+        omit_offsets : bool
             Omits offset terms in the `InferenceData` object returned when the model includes
             group specific effects.
-        include_mean: bool
+        include_mean : bool
             Compute the posterior of the mean response.
 
         Returns
@@ -386,6 +386,14 @@ def _run_laplace(self, draws, omit_offsets, include_mean):
         with self.model:
             maps = pm.find_MAP()
             n_maps = deepcopy(maps)
+
+            # Remove deterministics for parent parameters
+            n_maps = {
+                key: value
+                for key, value in n_maps.items()
+                if key not in self.spec.family.likelihood.params
+            }
+
             for m in maps:
                 if pm.util.is_transformed_name(m):
                     n_maps.pop(pm.util.get_untransformed_name(m))
@@ -398,6 +406,9 @@ def _run_laplace(self, draws, omit_offsets, include_mean):
         cov = np.linalg.inv(hessian)
         modes = np.concatenate([np.atleast_1d(v) for v in n_maps.values()])
 
+        print(cov.shape)
+        print(modes.shape)
+
         samples = np.random.multivariate_normal(modes, cov, size=draws)
 
         idata = _posterior_samples_to_idata(samples, self.model)
@@ -422,9 +433,9 @@ def _posterior_samples_to_idata(samples, model):
 
     Parameters
     ----------
-    samples: array
+    samples : array
         Posterior samples
-    model: PyMC model
+    model : PyMC model
 
     Returns
     -------

bambi/backend/terms.py

@@ -187,6 +187,9 @@ def build(self, spec):
             dist = dist(label, dims=dims, **self.term.prior.args)[np.newaxis, :]
         else:
             dist = dist(label, **self.term.prior.args)
+            # TODO: check if this brings the desired result
+            # Multiply it by vector of ones so it then has the proper length
+            dist = dist * np.ones((spec.response_component.term.data.shape[0],))
         return dist
 
     @property

bambi/families/family.py

@@ -226,7 +226,7 @@ def _make_dist_kwargs_and_coords(self, model, posterior, **kwargs):
         * A dictionary that maps the names of the likelihood parameters to draws from the
         posterior distribtuion.
         * An `xr.Coordinates` object with the coordinates required for the response. For example:
-        `(chain, draw, y_obs)` or `(chain, draw, y_obs, y_dim)`.
+        `(chain, draw, __obs__)` or `(chain, draw, __obs__, y_dim)`.
 
         It was created to abstract repetitive logic used in both `.posterior_predictive()` and
         `log_likelihood()`.

bambi/interpret/effects.py

@@ -35,6 +35,8 @@ class ResponseInfo:
     computing uncertainty intervals, and creating the summary dataframe in
     'PredictiveDifferences'.
 
+    FIXME: is the documentation outdated?
+
     Parameters
     ----------
     model : Model
@@ -65,7 +67,9 @@ def __post_init__(self):
         if self.target is None:
             self.name_target = self.name
         else:
-            self.name_target = f"{self.name}_{self.target}"
+            # TODO: verify this is the correct update
+            # self.name_target = f"{self.name}_{self.target}"
+            self.name_target = self.target
 
         self.name_obs = "__obs__"
         self.lower_bound_name = f"lower_{self.lower_bound * 100}%"
@@ -277,6 +281,9 @@ def get_estimate(
 
         if self.variable.values.ndim == 1:
             self.variable.values = np.array(self.variable.values).reshape(-1, 1)
+
+        print("name_target", self.response.name_target)
+        print("name_obs", self.response.name_obs)
 
         draws = {}
         variable_data = {}
@@ -515,11 +522,15 @@ def predictions(
             target = None
         else:
             # use the default response "y" and append target
-            response_name = get_aliased_name(model.response_component.response_term)
+            response_name = get_aliased_name(model.response_component.term)
     else:
-        response_name = get_aliased_name(model.response_component.response_term)
+        response_name = get_aliased_name(model.response_component.term)
 
-    response = ResponseInfo(response_name, target)
+    if target == "mean":
+        target_ = model.family.likelihood.parent
+    else:
+        target_ = target
+    response = ResponseInfo(response_name, target_)
     response_transform = transforms.get(response_name, identity)
 
     if pps:
@@ -680,10 +691,18 @@ def comparisons(
     conditional_info = ConditionalInfo(model, conditional)
 
     transforms = transforms if transforms is not None else {}
-    response_name = get_aliased_name(model.response_component.response_term)
+    response_name = get_aliased_name(model.response_component.term)
+
+    # TODO: see if this is appropriate 
     response = ResponseInfo(
-        response_name, target="mean", lower_bound=lower_bound, upper_bound=upper_bound
+        response_name, 
+        target=model.family.likelihood.parent, 
+        lower_bound=lower_bound, 
+        upper_bound=upper_bound
     )
+    # response = ResponseInfo(
+    #     response_name, target="mean", lower_bound=lower_bound, upper_bound=upper_bound
+    # )
     response_transform = transforms.get(response_name, identity)
 
     comparisons_data = create_differences_data(
@@ -844,7 +863,7 @@ def slopes(
     upper_bound = 1 - lower_bound
 
     transforms = transforms if transforms is not None else {}
-    response_name = get_aliased_name(model.response_component.response_term)
+    response_name = get_aliased_name(model.response_component.term)
     response = ResponseInfo(response_name, "mean", lower_bound, upper_bound)
     response_transform = transforms.get(response_name, identity)
 

bambi/interpret/plotting.py

@@ -44,7 +44,7 @@ def _plot_differences(
     if transforms is None:
         transforms = {}
 
-    response_name = get_aliased_name(model.response_component.response_term)
+    response_name = get_aliased_name(model.response_component.term)
 
     if ax is None:
         fig_kwargs = {} if fig_kwargs is None else fig_kwargs
@@ -73,7 +73,7 @@ def _plot_differences(
     else:
         raise TypeError("Main covariate must be numeric or categoric.")
 
-    response_name = get_aliased_name(model.response_component.response_term)
+    response_name = get_aliased_name(model.response_component.term)
     for ax in axes.ravel():  # pylint: disable = redefined-argument-from-local
         ax.set(xlabel=covariates.main, ylabel=response_name)
 
@@ -111,37 +111,37 @@ def plot_predictions(
     average_by: str, list, bool, optional
         The covariates we would like to average by. The passed covariate(s) will marginalize
         over the other covariates in the model. If True, it averages over all covariates
-        in the model to obtain the average estimate. Defaults to ``None``.
+        in the model to obtain the average estimate. Defaults to `None`.
     target : str
         Which model parameter to plot. Defaults to 'mean'. Passing a parameter into target only
         works when pps is False as the target may not be available in the posterior predictive
         distribution.
     sample_new_groups : bool, optional
         If the model contains group-level effects, and data is passed for unseen groups, whether
-        to sample from the new groups. Defaults to ``False``.
+        to sample from the new groups. Defaults to `False`.
     pps: bool, optional
-        Whether to plot the posterior predictive samples. Defaults to ``False``.
+        Whether to plot the posterior predictive samples. Defaults to `False`.
     use_hdi : bool, optional
         Whether to compute the highest density interval (defaults to True) or the quantiles.
     prob : float, optional
         The probability for the credibility intervals. Must be between 0 and 1. Defaults to 0.94.
-        Changing the global variable ``az.rcParam["stats.hdi_prob"]`` affects this default.
+        Changing the global variable `az.rcParam["stats.hdi_prob"]` affects this default.
     legend : bool, optional
-        Whether to automatically include a legend in the plot. Defaults to ``True``.
+        Whether to automatically include a legend in the plot. Defaults to `True`.
     transforms : dict, optional
         Transformations that are applied to each of the variables being plotted. The keys are the
-        name of the variables, and the values are functions to be applied. Defaults to ``None``.
+        name of the variables, and the values are functions to be applied. Defaults to `None`.
     ax : matplotlib.axes._subplots.AxesSubplot, optional
         A matplotlib axes object or a sequence of them. If None, this function instantiates a
-        new axes object. Defaults to ``None``.
+        new axes object. Defaults to `None`.
     fig_kwargs : optional
         Keyword arguments passed to the matplotlib figure function as a dict. For example,
-        ``fig_kwargs=dict(figsize=(11, 8)), sharey=True`` would make the figure 11 inches wide
+        `fig_kwargs=dict(figsize=(11, 8)), sharey=True` would make the figure 11 inches wide
         by 8 inches high and would share the y-axis values.
     subplot_kwargs : optional
         Keyword arguments used to determine the covariates used for the horizontal, group,
-        and panel axes. For example, ``subplot_kwargs=dict(main="x", group="y", panel="z")`` would
-        plot the horizontal axis as ``x``, the color (hue) as ``y``, and the panel axis as ``z``.
+        and panel axes. For example, `subplot_kwargs=dict(main="x", group="y", panel="z")` would
+        plot the horizontal axis as `x`, the color (hue) as `y`, and the panel axis as `z`.
 
     Returns
     -------
@@ -151,8 +151,8 @@ def plot_predictions(
     Raises
     ------
     ValueError
-        If ``conditional`` and ``average_by`` are both ``None``.
-        If length of ``conditional`` is greater than 3 and ``average_by`` is ``None``.
+        If `conditional` and `average_by` are both `None`.
+        If length of `conditional` is greater than 3 and `average_by` is `None`.
         If main covariate is not numeric or categoric.
     """
     if conditional is None and average_by is None:
@@ -207,7 +207,7 @@ def plot_predictions(
     else:
         covariates = get_covariates(conditional_info.covariates)
 
-    response_name = get_aliased_name(model.response_component.response_term)
+    response_name = get_aliased_name(model.response_component.term)
 
     if ax is None:
         fig_kwargs = {} if fig_kwargs is None else fig_kwargs
@@ -270,33 +270,33 @@ def plot_comparisons(
         values are the values to condition on.
     average_by: str, list, optional
         The covariates we would like to average by. The passed covariate(s) will marginalize
-        over the other covariates in the model. Defaults to ``None``.
+        over the other covariates in the model. Defaults to `None`.
     comparison_type : str, optional
         The type of comparison to plot. Defaults to 'diff'.
     sample_new_groups : bool, optional
         If the model contains group-level effects, and data is passed for unseen groups, whether
-        to sample from the new groups. Defaults to ``False``.
+        to sample from the new groups. Defaults to `False`.
     use_hdi : bool, optional
         Whether to compute the highest density interval (defaults to True) or the quantiles.
     prob : float, optional
         The probability for the credibility intervals. Must be between 0 and 1. Defaults to 0.94.
-        Changing the global variable ``az.rcParam["stats.hdi_prob"]`` affects this default.
+        Changing the global variable `az.rcParam["stats.hdi_prob"]` affects this default.
     legend : bool, optional
-        Whether to automatically include a legend in the plot. Defaults to ``True``.
+        Whether to automatically include a legend in the plot. Defaults to `True`.
     transforms : dict, optional
         Transformations that are applied to each of the variables being plotted. The keys are the
-        name of the variables, and the values are functions to be applied. Defaults to ``None``.
+        name of the variables, and the values are functions to be applied. Defaults to `None`.
     ax : matplotlib.axes._subplots.AxesSubplot, optional
         A matplotlib axes object or a sequence of them. If None, this function instantiates a
-        new axes object. Defaults to ``None``.
+        new axes object. Defaults to `None`.
     fig_kwargs : optional
         Keyword arguments passed to the matplotlib figure function as a dict. For example,
-        ``fig_kwargs=dict(figsize=(11, 8)), sharey=True`` would make the figure 11 inches wide
+        `fig_kwargs=dict(figsize=(11, 8)), sharey=True` would make the figure 11 inches wide
         by 8 inches high and would share the y-axis values.
     subplot_kwargs : optional
         Keyword arguments used to determine the covariates used for the horizontal, group,
-        and panel axes. For example, ``subplot_kwargs=dict(main="x", group="y", panel="z")`` would
-        plot the horizontal axis as ``x``, the color (hue) as ``y``, and the panel axis as ``z``.
+        and panel axes. For example, `subplot_kwargs=dict(main="x", group="y", panel="z")` would
+        plot the horizontal axis as `x`, the color (hue) as `y`, and the panel axis as `z`.
 
     Returns
     -------
@@ -306,10 +306,10 @@ def plot_comparisons(
     Raises
     ------
     ValueError
-        If the number of contrast levels is greater than 2 and ``average_by`` is ``None``.
-        If ``conditional`` and ``average_by`` are both ``None``.
-        If length of ``conditional`` is greater than 3 and ``average_by`` is ``None``.
-        If ``average_by`` is ``True``.
+        If the number of contrast levels is greater than 2 and `average_by` is `None`.
+        If `conditional` and `average_by` are both `None`.
+        If length of `conditional` is greater than 3 and `average_by` is `None`.
+        If `average_by` is `True`.
         If main covariate is not numeric or categoric.
     """
     contrast_name = contrast
@@ -416,7 +416,7 @@ def plot_slopes(
     average_by: str, list, bool, optional
         The covariates we would like to average by. The passed covariate(s) will marginalize
         over the other covariates in the model. If True, it averages over all covariates
-        in the model to obtain the average estimate. Defaults to ``None``.
+        in the model to obtain the average estimate. Defaults to `None`.
     eps : float, optional
         To compute the slope, 'wrt' is evaluated at wrt +/- 'eps'. The rate of change is then
         computed as the difference between the two values divided by 'eps'. Defaults to 1e-4.
@@ -432,28 +432,28 @@ def plot_slopes(
         in the response.
     sample_new_groups : bool, optional
         If the model contains group-level effects, and data is passed for unseen groups, whether
-        to sample from the new groups. Defaults to ``False``.
+        to sample from the new groups. Defaults to `False`.
     use_hdi : bool, optional
         Whether to compute the highest density interval (defaults to True) or the quantiles.
     prob : float, optional
         The probability for the credibility intervals. Must be between 0 and 1. Defaults to 0.94.
-        Changing the global variable ``az.rcParam["stats.hdi_prob"]`` affects this default.
+        Changing the global variable `az.rcParam["stats.hdi_prob"]` affects this default.
     transforms : dict, optional
         Transformations that are applied to each of the variables being plotted. The keys are the
-        name of the variables, and the values are functions to be applied. Defaults to ``None``.
+        name of the variables, and the values are functions to be applied. Defaults to `None`.
     legend : bool, optional
-        Whether to automatically include a legend in the plot. Defaults to ``True``.
+        Whether to automatically include a legend in the plot. Defaults to `True`.
     ax : matplotlib.axes._subplots.AxesSubplot, optional
         A matplotlib axes object or a sequence of them. If None, this function instantiates a
-        new axes object. Defaults to ``None``.
+        new axes object. Defaults to `None`.
     fig_kwargs : optional
         Keyword arguments passed to the matplotlib figure function as a dict. For example,
-        ``fig_kwargs=dict(figsize=(11, 8)), sharey=True`` would make the figure 11 inches wide
+        `fig_kwargs=dict(figsize=(11, 8)), sharey=True` would make the figure 11 inches wide
         by 8 inches high and would share the y-axis values.
     subplot_kwargs : optional
         Keyword arguments used to determine the covariates used for the horizontal, group,
-        and panel axes. For example, ``subplot_kwargs=dict(main="x", group="y", panel="z")`` would
-        plot the horizontal axis as ``x``, the color (hue) as ``y``, and the panel axis as ``z``.
+        and panel axes. For example, `subplot_kwargs=dict(main="x", group="y", panel="z")` would
+        plot the horizontal axis as `x`, the color (hue) as `y`, and the panel axis as `z`.
 
     Returns
     -------
@@ -463,11 +463,11 @@ def plot_slopes(
     Raises
     ------
     ValueError
-        If the number of ``wrt`` values is greater than 2 and ``average_by`` is ``None``.
-        If ``conditional`` and ``average_by`` are both ``None``.
-        If length of ``conditional`` is greater than 3 and ``average_by`` is ``None``.
-        If ``average_by`` is ``True``.
-        If ``slope`` is not one of ('dydx', 'dyex', 'eyex', 'eydx').
+        If the number of `wrt` values is greater than 2 and `average_by` is `None`.
+        If `conditional` and `average_by` are both `None`.
+        If length of `conditional` is greater than 3 and `average_by` is `None`.
+        If `average_by` is `True`.
+        If `slope` is not one of ('dydx', 'dyex', 'eyex', 'eydx').
         If main covariate is not numeric or categoric.
     """
     wrt_name = wrt

tests/test_alternative_samplers.py

@@ -29,7 +29,6 @@ def data_n100():
     )
     return data
 
-
 def test_laplace():
     data = pd.DataFrame(np.repeat((0, 1), (30, 60)), columns=["w"])
     priors = {"Intercept": bmb.Prior("Uniform", lower=0, upper=1)}