Adds functions ceil, round to numpy API for onnx

_doc/sphinxdoc/source/api/npy.rst

@@ -97,6 +97,8 @@ Available numpy functions implemented with ONNX operators
 
 .. autosignature:: mlprodict.npy.numpy_onnx_impl.atanh
 
+.. autosignature:: mlprodict.npy.numpy_onnx_impl.ceil
+
 .. autosignature:: mlprodict.npy.numpy_onnx_impl.cos
 
 .. autosignature:: mlprodict.npy.numpy_onnx_impl.cosh
@@ -117,6 +119,8 @@ Available numpy functions implemented with ONNX operators
 
 .. autosignature:: mlprodict.npy.numpy_onnx_impl.relu
 
+.. autosignature:: mlprodict.npy.numpy_onnx_impl.round
+
 .. autosignature:: mlprodict.npy.numpy_onnx_impl.sign
 
 .. autosignature:: mlprodict.npy.numpy_onnx_impl.sin
@@ -156,6 +160,8 @@ ONNX functions executed python ONNX runtime
 
 .. autosignature:: mlprodict.npy.numpy_onnx_pyrt.atanh
 
+.. autosignature:: mlprodict.npy.numpy_onnx_pyrt.ceil
+
 .. autosignature:: mlprodict.npy.numpy_onnx_pyrt.cos
 
 .. autosignature:: mlprodict.npy.numpy_onnx_pyrt.cosh
@@ -176,6 +182,8 @@ ONNX functions executed python ONNX runtime
 
 .. autosignature:: mlprodict.npy.numpy_onnx_pyrt.relu
 
+.. autosignature:: mlprodict.npy.numpy_onnx_pyrt.round
+
 .. autosignature:: mlprodict.npy.numpy_onnx_pyrt.sign
 
 .. autosignature:: mlprodict.npy.numpy_onnx_pyrt.sin

_unittests/ut_npy/test_numpy_onnx_pyrt.py

@@ -100,6 +100,10 @@ def test_atanh_float32(self):
         x = numpy.array([[0.5, 0.1], [-0.5, -0.1]], dtype=numpy.float32)
         self.common_test1(x, numpy.arctanh, nxnpy.atanh, numpy.float32)
 
+    def test_ceil_float32(self):
+        x = numpy.array([[0.5, 0.1], [-0.5, -0.1]], dtype=numpy.float32)
+        self.common_test1(x, numpy.ceil, nxnpy.ceil, numpy.float32)
+
     def test_cos_float32(self):
         x = numpy.array([[0.5, 0.1], [-0.5, -0.1]], dtype=numpy.float32)
         self.common_test1(x, numpy.cos, nxnpy.cos, numpy.float32)
@@ -152,6 +156,10 @@ def test_relu_float64(self):
         self.common_test1(x, lambda x: numpy.maximum(x, 0),
                           nxnpy.relu, numpy.float64)
 
+    def test_round_float64(self):
+        x = numpy.array([[6.1, 5], [3.5, -7.8]], dtype=numpy.float64)
+        self.common_test1(x, numpy.round, nxnpy.round, numpy.float64)
+
     def test_sign_float64(self):
         x = numpy.array([[-6.1, 5], [3.5, 7.8]], dtype=numpy.float64)
         self.common_test1(x, numpy.sign, nxnpy.sign, numpy.float64)

_unittests/ut_npy/test_onnx_variable_ort.py

@@ -477,12 +477,12 @@ def test_ort_abs_set3(self):
     def test_ort_log(self):
         x = numpy.array([[6.1, 5], [3.5, 7.8]], dtype=numpy.float32)
         y = test_log(x)
-        self.assertEqualArray(y, numpy.log(x))
+        self.assertEqualArray(y, numpy.log(x), decimal=6)
 
     def test_ort_abs_log_multi(self):
         x = numpy.array([[6.1, -5], [-3.5, 7.8]], dtype=numpy.float32)
         y = test_abs_log_multi(x)
-        self.assertEqualArray(y, numpy.log(numpy.abs(x)))
+        self.assertEqualArray(y, numpy.log(numpy.abs(x)), decimal=6)
 
     def test_ort_abs_shape(self):
         x = numpy.array([[6.1, -5], [3.5, -7.8]], dtype=numpy.float32)

_unittests/ut_onnxrt/test_onnxrt_python_runtime_.py

@@ -51,6 +51,7 @@
     OnnxReduceSum, OnnxReduceSumApi11, OnnxReduceSum_11, OnnxReduceSum_1,
     OnnxReduceSumSquare,
     OnnxRelu, OnnxReshape,
+    OnnxRound,
     OnnxScatterElements, OnnxShape, OnnxSlice, OnnxSigmoid, OnnxSign,
     OnnxSin, OnnxSinh,
     OnnxSize, OnnxSoftmax,
@@ -2335,6 +2336,10 @@ def test_onnxt_runtime_relu(self):
         self.common_test_onnxt_runtime_unary(
             OnnxRelu, lambda x: numpy.maximum(x, 0))
 
+    @wraplog()
+    def test_onnxt_runtime_round(self):
+        self.common_test_onnxt_runtime_unary(OnnxRound, numpy.round)
+
     @ignore_warnings(category=(RuntimeWarning, DeprecationWarning))
     @wraplog()
     def test_onnxt_runtime_reshape(self):

mlprodict/npy/numpy_onnx_impl.py

@@ -12,6 +12,7 @@
     OnnxArgMin,
     OnnxAsin, OnnxAsinh,
     OnnxAtan, OnnxAtanh,
+    OnnxCeil,
     OnnxCos, OnnxCosh,
     OnnxErf,
     OnnxExp,
@@ -24,6 +25,7 @@
     OnnxReduceProd,
     OnnxReduceSum,
     OnnxRelu,
+    OnnxRound,
     OnnxSign,
     OnnxSin, OnnxSinh,
     OnnxSqrt,
@@ -99,6 +101,11 @@ def atanh(x):
     return OnnxVar(x, op=OnnxAtanh)
 
 
+def ceil(x):
+    "See :epkg:`numpy:ceil`."
+    return OnnxVar(x, op=OnnxCeil)
+
+
 def cos(x):
     "See :epkg:`numpy:cos`."
     return OnnxVar(x, op=OnnxCos)
@@ -157,6 +164,11 @@ def reciprocal(x):
     return OnnxVar(x, op=OnnxReciprocal)
 
 
+def round(x):
+    "See :epkg:`numpy:round`."
+    return OnnxVar(x, op=OnnxRound)
+
+
 def sign(x):
     "See :epkg:`numpy:sign`."
     return OnnxVar(x, op=OnnxSign)

mlprodict/npy/numpy_onnx_pyrt.py

@@ -20,6 +20,7 @@
     asinh as nx_asinh,
     atan as nx_atan,
     atanh as nx_atanh,
+    ceil as nx_ceil,
     cos as nx_cos,
     cosh as nx_cosh,
     erf as nx_erf,
@@ -30,6 +31,7 @@
     prod as nx_prod,
     reciprocal as nx_reciprocal,
     relu as nx_relu,
+    round as nx_round,
     sign as nx_sign,
     sin as nx_sin,
     sinh as nx_sinh,
@@ -107,6 +109,12 @@ def atanh(x):
     return nx_atanh(x)
 
 
+@onnxnumpy_np(signature=NDArraySameTypeSameShape("floats"))
+def ceil(x):
+    "ceil"
+    return nx_ceil(x)
+
+
 @onnxnumpy_np(signature=NDArraySameTypeSameShape("floats"))
 def cos(x):
     "cos"
@@ -167,6 +175,12 @@ def relu(x):
     return nx_relu(x)
 
 
+@onnxnumpy_np(signature=NDArraySameTypeSameShape("floats"))
+def round(x):
+    "round"
+    return nx_round(x)
+
+
 @onnxnumpy_np(signature=NDArraySameTypeSameShape("floats"))
 def sign(x):
     "sign"

mlprodict/onnxrt/ops_cpu/_op_list.py

@@ -87,6 +87,7 @@
 from .op_reduce_sum_square import ReduceSumSquare
 from .op_relu import Relu
 from .op_reshape import Reshape, Reshape_5, Reshape_13, Reshape_14
+from .op_round import Round
 from .op_rnn import RNN
 from .op_scaler import Scaler
 from .op_scan import Scan

mlprodict/onnxrt/ops_cpu/op_round.py

@@ -0,0 +1,26 @@
+# -*- encoding: utf-8 -*-
+# pylint: disable=E0203,E1101,C0111
+"""
+@file
+@brief Runtime operator.
+"""
+import numpy
+from ._op import OpRunUnaryNum
+
+
+class Round(OpRunUnaryNum):
+
+    def __init__(self, onnx_node, desc=None, **options):
+        OpRunUnaryNum.__init__(self, onnx_node, desc=desc,
+                               **options)
+
+    def _run(self, x):  # pylint: disable=W0221
+        if self.inplaces.get(0, False):
+            return self._run_inplace(x)
+        return (numpy.round(x), )
+
+    def _run_inplace(self, x):
+        return (numpy.round(x, out=x), )
+
+    def to_python(self, inputs):
+        return self._to_python_numpy(inputs, self.__class__.__name__.lower())