abcpp 1.0.0__cp313-cp313-win_amd64.whl

abcpp/__init__.py

@@ -0,0 +1,4 @@
+from .abc import abc
+from .summary import summary
+
+__all__ = ["abc", "summary"]

abcpp/abc.py

@@ -0,0 +1,120 @@
+import copy
+
+import numpy
+
+from . import _core
+
+
+DEFAULT_CONTROL = {
+    "method": "rejection",
+    "tol": 0.01,
+    "kernel": "epanechnikov",
+    "hcorr": True,
+    "transf": "none",
+    "logit_bounds": None,
+    "subset": None,
+    "prior_weights": None,
+    "seed": 1004,
+    "reduction": "none",
+    "n_comp": 0,
+    "nnet": {
+        "numnet": 10,
+        "sizenet": 5,
+        "lambda": [0.0001, 0.001, 0.01],
+        "maxit": 500,
+        "rang": 0.7,
+        "abstol": 1e-4,
+        "reltol": 1e-8,
+        "verbose": False,
+        "skip": False,
+    },
+}
+
+
+def _nested_merge(defaults, overrides):
+    merged = copy.deepcopy(defaults)
+    for key, value in overrides.items():
+        if (
+            isinstance(value, dict)
+            and isinstance(merged.get(key), dict)
+        ):
+            merged[key] = _nested_merge(merged[key], value)
+        else:
+            merged[key] = value
+    return merged
+
+
+def _prepare_control(control):
+    if control is None:
+        control = {}
+    if not isinstance(control, dict):
+        raise TypeError("control must be a dict or None")
+    merged = _nested_merge(DEFAULT_CONTROL, control)
+    if isinstance(merged["transf"], str):
+        merged["transf"] = [merged["transf"]]
+    if merged["logit_bounds"] is None:
+        merged["logit_bounds"] = numpy.zeros((1, 2), dtype=float)
+    if merged["subset"] is None:
+        merged["subset"] = numpy.asarray([], dtype=bool)
+    if merged["prior_weights"] is None:
+        merged["prior_weights"] = numpy.asarray([], dtype=float)
+    else:
+        merged["prior_weights"] = numpy.asarray(
+            merged["prior_weights"],
+            dtype=float,
+        )
+    return merged
+
+
+def _is_matrix_like(value):
+    try:
+        return numpy.asarray(value).ndim == 2
+    except ValueError:
+        return False
+
+
+def abc(target, params, sumstats, control=None):
+    """Run Approximate Bayesian Computation with the C++ backend."""
+    control = _prepare_control(control)
+    target_array = numpy.asarray(target, dtype=float)
+    param_array = numpy.asarray(params, dtype=float)
+    sumstat_is_mapping = isinstance(sumstats, dict)
+    sumstat_items = list(sumstats.values()) if sumstat_is_mapping else sumstats
+    sumstat_is_matrix_collection = (
+        isinstance(sumstat_items, (list, tuple))
+        and len(sumstat_items) > 0
+        and all(_is_matrix_like(item) for item in sumstat_items)
+    )
+    if sumstat_is_matrix_collection:
+        sumstat_value = [
+            numpy.asarray(item, dtype=float)
+            for item in sumstat_items
+        ]
+    else:
+        sumstat_value = numpy.asarray(sumstats, dtype=float)
+
+    if target_array.ndim == 1:
+        target_array = target_array.reshape((1, -1))
+    if param_array.ndim == 1:
+        param_array = param_array.reshape((-1, 1))
+    if not sumstat_is_matrix_collection and sumstat_value.ndim == 1:
+        sumstat_value = sumstat_value.reshape((-1, 1))
+    if (
+        not sumstat_is_matrix_collection
+        and param_array.ndim == 2
+        and sumstat_value.ndim == 2
+        and param_array.shape[0] == 1
+        and param_array.shape[1] == sumstat_value.shape[0]
+        and sumstat_value.shape[0] != 1
+    ):
+        param_array = param_array.T
+
+    if sumstat_is_matrix_collection:
+        return _core.abc_matrix_list(
+            target_array,
+            param_array,
+            [numpy.asarray(item, dtype=float) for item in sumstat_value],
+            control,
+        )
+
+    return _core.abc(target_array, param_array, sumstat_value, control)

abcpp/summary.py

@@ -0,0 +1,6 @@
+from . import _core
+
+
+def summary(result, unadj=False, intvl=0.95):
+    """Summarize posterior draws returned by :func:`abcpp.abc`."""
+    return _core.summary(result, unadj=bool(unadj), intvl=float(intvl))