ufunc removal

.github/workflows/array_api.yml

@@ -55,6 +55,6 @@ jobs:
           # to circumvent the currently slow performance of
           # JIT compile/link, which can otherwise cause issues
           # for hypothesis-driven test case generation
-          pytest $GITHUB_WORKSPACE/tools/pre_compile_ufuncs.py -s
+          # pytest $GITHUB_WORKSPACE/tools/pre_compile_ufuncs.py -s
           # only run a subset of the conformance tests to get started
           pytest array_api_tests/meta/test_broadcasting.py array_api_tests/meta/test_equality_mapping.py array_api_tests/meta/test_signatures.py array_api_tests/meta/test_special_cases.py array_api_tests/test_constants.py array_api_tests/meta/test_utils.py array_api_tests/test_creation_functions.py::test_ones array_api_tests/test_creation_functions.py::test_ones_like array_api_tests/test_data_type_functions.py::test_result_type array_api_tests/test_operators_and_elementwise_functions.py::test_log10 array_api_tests/test_operators_and_elementwise_functions.py::test_sqrt array_api_tests/test_operators_and_elementwise_functions.py::test_isfinite array_api_tests/test_operators_and_elementwise_functions.py::test_log2 array_api_tests/test_operators_and_elementwise_functions.py::test_log1p array_api_tests/test_operators_and_elementwise_functions.py::test_isinf array_api_tests/test_operators_and_elementwise_functions.py::test_log array_api_tests/test_array_object.py::test_scalar_casting array_api_tests/test_operators_and_elementwise_functions.py::test_sign array_api_tests/test_operators_and_elementwise_functions.py::test_square array_api_tests/test_operators_and_elementwise_functions.py::test_cos array_api_tests/test_operators_and_elementwise_functions.py::test_round array_api_tests/test_operators_and_elementwise_functions.py::test_trunc array_api_tests/test_operators_and_elementwise_functions.py::test_ceil array_api_tests/test_operators_and_elementwise_functions.py::test_floor array_api_tests/test_operators_and_elementwise_functions.py::test_exp array_api_tests/test_operators_and_elementwise_functions.py::test_sin array_api_tests/test_operators_and_elementwise_functions.py::test_tan array_api_tests/test_operators_and_elementwise_functions.py::test_tanh array_api_tests/test_creation_functions.py::test_zeros array_api_tests/test_creation_functions.py::test_zeros_like array_api_tests/test_creation_functions.py::test_full_like array_api_tests/test_operators_and_elementwise_functions.py::test_positive array_api_tests/test_operators_and_elementwise_functions.py::test_isnan array_api_tests/test_operators_and_elementwise_functions.py::test_equal "array_api_tests/test_has_names.py::test_has_names[array_method-__pos__]"

examples/LogisticRegression/LR.py

@@ -273,7 +273,7 @@ def _logistic_regression_path(
 
     _, n_features = X.shape
 
-    classes = pk.unique(y)
+    classes = np.unique(y)
 
     random_state = check_random_state(random_state)
 
@@ -333,8 +333,8 @@ def _logistic_regression_path(
             lbin = LabelBinarizer()
             Y_multi = asarray(lbin.fit_transform(y))
             if Y_multi.shape[1] == 1:
-                Y_multi = pk.hstack(
-                    pk.negative(pk.subtract(Y_multi, asarray([1]))), Y_multi
+                Y_multi = np.hstack(
+                    np.negative(np.subtract(Y_multi, asarray([1]))), Y_multi
                 )
 
         w0 = pk.zeros((classes.size, n_features + int(fit_intercept)), dtype=X.dtype)
@@ -397,7 +397,7 @@ def _logistic_regression_path(
             func = loss.loss
             grad = loss.gradient
             hess = loss.gradient_hessian_product  # hess = [gradient, hessp]
-        warm_start_sag = {"coef": pk.transpose(w0)}
+        warm_start_sag = {"coef": np.transpose(np.array(w0))}
     else:
         target = y_bin
         if solver == "lbfgs":
@@ -471,7 +471,7 @@ def _logistic_regression_path(
             )
             coef_ = asarray(coef_)
             if fit_intercept:
-                w0 = pk.hstack(pk.ravel(coef_), intercept_)
+                w0 = np.hstack(pk.ravel(coef_), intercept_)
             else:
                 w0 = pk.ravel(coef_)
 
@@ -862,7 +862,7 @@ def fit(self, X, y, sample_weight=None):
 
         X = asarray(X)
         y = asarray(y)
-        self.classes_ = pk.unique(y)
+        self.classes_ = np.unique(np.array(y))
 
         multi_class = _check_multi_class(self.multi_class, solver, len(self.classes_))
 
@@ -1024,7 +1024,7 @@ def predict_proba(self, X):
             if decision.ndim == 1:
                 # Workaround for multi_class="multinomial" and binary outcomes
                 # which requires softmax prediction with only a 1D decision.
-                decision_2d = pk.hstack(pk.negative(decision), decision)
+                decision_2d = np.hstack((np.negative(decision), decision))
             else:
                 decision_2d = decision
             return softmax(decision_2d, copy=False)
@@ -1045,7 +1045,7 @@ def predict_log_proba(self, X):
             Returns the log-probability of the sample for each class in the
             model, where classes are ordered as they are in ``self.classes_``.
         """
-        return pk.log(self.predict_proba(X))
+        return np.log(self.predict_proba(X))
 
     def predict(self, X):
         """
@@ -1065,7 +1065,7 @@ def predict(self, X):
         else:
             indices = scores.argmax(axis=1)
 
-        return pk.index(self.classes_, asarray(indices, dtype=pk.int32))
+        return self.classes_[np.array(indices, dtype=np.int32)]
 
 
 def main():

examples/NaiveBayes/GaussianNB.py

@@ -95,15 +95,15 @@ def type_of_target(y, input_name=""):
     else:
         suffix = ""  # [1, 2, 3] or [[1], [2], [3]]
 
-    if (len(pk.unique(y)) > 2) or (len(y.shape) >= 2 and len(y[0]) > 1):
+    if (len(np.unique(y)) > 2) or (len(y.shape) >= 2 and len(y[0]) > 1):
         return "multiclass" + suffix  # [1, 2, 3] or [[1., 2., 3]] or [[1, 2]]
     else:
         return "binary"  # [1, 2] or [["a"], ["b"]]
 
 
 def _unique_multiclass(y):
     if hasattr(y, "__array__"):
-        return pk.unique(asarray(y))
+        return np.unique(asarray(y))
     else:
         return set(y)
 
@@ -264,7 +264,7 @@ def predict(self, X):
         X = self._check_X(X)
         jll = self._joint_log_likelihood(X)
 
-        return pk.index(self.classes_, pk.argmax(jll, axis=1))
+        return self.classes_[np.argmax(np.array(jll), axis=1)]
 
     def predict_log_proba(self, X):
         """
@@ -285,7 +285,7 @@ def predict_log_proba(self, X):
         jll = self._joint_log_likelihood(X)
         # normalize by P(x) = P(f_1, ..., f_n)
         # log_prob_x = logsumexp(jll, axis=1)
-        # return jll - pk.transpose(pk.atleast_2d())
+        # return jll - np.transpose(pk.atleast_2d())
 
     def predict_proba(self, X):
         """
@@ -301,7 +301,7 @@ def predict_proba(self, X):
             the model. The columns correspond to the classes in sorted
             order, as they appear in the attribute :term:`classes_`.
         """
-        return pk.exp(self.predict_log_proba(X))
+        return np.exp(self.predict_log_proba(X))
 
 
 class GaussianNB(_BaseNB):
@@ -366,7 +366,7 @@ class labels known to the classifier.
     >>> print(clf.predict([[-0.8, -1]]))
     [1]
     >>> clf_pf = GaussianNB()
-    >>> clf_pf.partial_fit(X, Y, pk.unique(Y))
+    >>> clf_pf.partial_fit(X, Y, np.unique(Y))
     GaussianNB()
     >>> print(clf_pf.predict([[-0.8, -1]]))
     [1]
@@ -397,7 +397,7 @@ def fit(self, X, y, sample_weight=None):
         y = asarray(self._validate_data(y=y))
 
         return self._partial_fit(
-            X, y, pk.unique(y), _refit=True, sample_weight=sample_weight
+            X, y, np.unique(y), _refit=True, sample_weight=sample_weight
         )
 
     def _check_X(self, X):
@@ -440,12 +440,14 @@ def _update_mean_variance(n_past, mu, var, X, sample_weight=None):
         # Compute (potentially weighted) mean and variance of new datapoints
         if sample_weight is not None:
             n_new = float(sample_weight.sum())
-            new_mu = pk.average(X, axis=0, weights=sample_weight)
-            new_var = pk.average((X - new_mu) ** 2, axis=0, weights=sample_weight)
+            new_mu = np.average(np.array(X), axis=0, weights=sample_weight)
+            new_var = np.average(
+                np.array(X - new_mu) ** 2, axis=0, weights=sample_weight
+            )
         else:
             n_new = X.shape[0]
-            new_var = pk.var(X, axis=0)
-            new_mu = pk.mean(X, axis=0)
+            new_var = np.var(np.array(X), axis=0)
+            new_mu = np.mean(np.array(X), axis=0)
 
         if n_past == 0:
             return new_mu, new_var
@@ -534,7 +536,7 @@ def _partial_fit(self, X, y, classes=None, _refit=False, sample_weight=None):
         # will cause numerical errors. To address this, we artificially
         # boost the variance by epsilon, a small fraction of the standard
         # deviation of the largest dimension.
-        self.epsilon_ = self.var_smoothing * pk.find_max(pk.var(X, axis=0))
+        self.epsilon_ = self.var_smoothing * pk.find_max(np.var(np.array(X), axis=0))
 
         if first_call:
             # This is the first call to partial_fit:
@@ -569,13 +571,13 @@ def _partial_fit(self, X, y, classes=None, _refit=False, sample_weight=None):
 
         classes = self.classes_
 
-        unique_y = pk.unique(y)
-        unique_y_in_classes = pk.in1d(unique_y, classes)
+        unique_y = np.unique(y)
+        unique_y_in_classes = np.in1d(unique_y, classes)
 
         if not pk.all(unique_y_in_classes):
             raise ValueError(
                 "The target label(s) %s in y do not exist in the initial classes %s"
-                % (unique_y[pk.logical_not(unique_y_in_classes)], classes)
+                % (unique_y[np.logical_not(unique_y_in_classes)], classes)
             )
 
         for y_i in unique_y:
@@ -602,7 +604,7 @@ def _partial_fit(self, X, y, classes=None, _refit=False, sample_weight=None):
         # Update if only no priors is provided
         if self.priors is None:
             # Empirical prior, with sample_weight taken into account
-            self.class_prior_ = pk.divide(self.class_count_, pk.sum(self.class_count_))
+            self.class_prior_ = np.divide(self.class_count_, pk.sum(self.class_count_))
 
         return self
 
@@ -611,15 +613,15 @@ def _joint_log_likelihood(self, X):
         total_classes = reduce(lambda a, b: a * b, self.classes_.shape, 1)
 
         for i in range(total_classes):
-            jointi = pk.log(self.class_prior_[i])
+            jointi = np.log(self.class_prior_[i])
 
-            n_ij = -0.5 * pk.sum(pk.log(pk.multiply(self.var_[i, :], 2.0 * pi)))
-            n_ij = pk.add(
-                pk.negative(
-                    pk.multiply(
+            n_ij = -0.5 * pk.sum(np.log(np.multiply(self.var_[i, :], 2.0 * pi)))
+            n_ij = np.add(
+                np.negative(
+                    np.multiply(
                         pk.sum(
-                            pk.divide(
-                                pk.power(pk.add(X, pk.negative(self.theta_[i, :])), 2),
+                            np.divide(
+                                np.power(np.add(X, np.negative(self.theta_[i, :])), 2),
                                 self.var_[i, :],
                             ),
                             1,
@@ -630,9 +632,9 @@ def _joint_log_likelihood(self, X):
                 n_ij,
             )
 
-            joint_log_likelihood.append(pk.add(n_ij, jointi))
+            joint_log_likelihood.append(np.add(n_ij, jointi))
 
-        joint_log_likelihood = pk.transpose(asarray(joint_log_likelihood))
+        joint_log_likelihood = np.transpose(asarray(joint_log_likelihood))
         return joint_log_likelihood
 
 

pykokkos/__init__.py

@@ -17,62 +17,6 @@
     set_device_id,
 )
 
-from pykokkos.lib.ufuncs import (
-    reciprocal,
-    log,
-    log2,
-    log10,
-    log1p,
-    sqrt,
-    sign,
-    add,
-    copyto,
-    subtract,
-    dot,
-    multiply,
-    matmul,
-    np_matmul,
-    divide,
-    negative,
-    positive,
-    power,
-    fmod,
-    square,
-    greater,
-    logaddexp,
-    true_divide,
-    logaddexp2,
-    floor_divide,
-    sin,
-    cos,
-    tan,
-    tanh,
-    logical_and,
-    logical_or,
-    logical_xor,
-    logical_not,
-    fmax,
-    fmin,
-    exp,
-    exp2,
-    argmax,
-    unique,
-    var,
-    in1d,
-    mean,
-    hstack,
-    transpose,
-    index,
-    isinf,
-    isnan,
-    equal,
-    isfinite,
-    round,
-    trunc,
-    ceil,
-    floor,
-    broadcast_view,
-)
 from pykokkos.lib.info import iinfo, finfo
 from pykokkos.lib.create import zeros, zeros_like, ones, ones_like, full, full_like
 from pykokkos.lib.manipulate import reshape, ravel, expand_dims
@@ -89,10 +33,6 @@
 from pykokkos.lib.constants import e, pi, inf, nan
 from pykokkos.interface.views import astype
 
-__array_api_version__ = "2021.12"
-
-__all__ = ["__array_api_version__"]
-
 runtime_singleton.runtime = Runtime()
 
 import weakref

pykokkos/interface/views.py

@@ -503,8 +503,6 @@ def _get_type(self, dtype: Union[DataType, type]) -> Optional[DataType]:
         return None
 
     def __eq__(self, other):
-        # avoid circular import with scoped import
-        from pykokkos.lib.ufuncs import equal
 
         if isinstance(other, float):
             new_other = pk.View((), dtype=pk.double)
@@ -539,7 +537,10 @@ def __eq__(self, other):
             new_other = other
         else:
             raise ValueError("unexpected types!")
-        return equal(self, new_other)
+        result_np = np.equal(np.array(self), np.array(new_other))
+        result = pk.View(result_np.shape, dtype=pk.bool)
+        result[:] = result_np
+        return result
 
     def __hash__(self):
         try:
@@ -667,8 +668,6 @@ def _get_base_view(self, parent_view: Union[Subview, View]) -> View:
         return base_view
 
     def __eq__(self, other):
-        # avoid circular import with scoped import
-        from pykokkos.lib.ufuncs import equal
 
         if isinstance(other, float):
             new_other = pk.View((), dtype=pk.double)
@@ -703,7 +702,10 @@ def __eq__(self, other):
             new_other = other
         else:
             raise ValueError("unexpected types!")
-        return equal(self, new_other)
+        result_np = np.equal(np.array(self), np.array(new_other))
+        result = pk.View(result_np.shape, dtype=pk.bool)
+        result[:] = result_np
+        return result
 
     def __add__(self, other):
         if isinstance(other, float):