Add Erf operator for computing error function (onnx#1675)

* Add Erf operator for computing error function

* update test coverage

Author: bddppq

docs/Changelog.md

@@ -8648,6 +8648,35 @@ This version of the operator has been available since version 9 of the default O
 <dd>Constrain indices to integer types</dd>
 </dl>
 
+### <a name="Erf-9"></a>**Erf-9**</a>
+
+  Computes the error function of the given input tensor element-wise.
+
+#### Version
+
+This version of the operator has been available since version 9 of the default ONNX operator set.
+
+#### Inputs
+
+<dl>
+<dt><tt>input</tt> : T</dt>
+<dd>Input tensor</dd>
+</dl>
+
+#### Outputs
+
+<dl>
+<dt><tt>output</tt> : T</dt>
+<dd>The error function of the input tensor computed element-wise. It has the same shape and type of the input.</dd>
+</dl>
+
+#### Type Constraints
+
+<dl>
+<dt><tt>T</tt> : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)</dt>
+<dd>Constrain input and output types to all numeric tensors.</dd>
+</dl>
+
 ### <a name="EyeLike-9"></a>**EyeLike-9**</a>
 
   Generate a 2D tensor (matrix) with ones on the diagonal and zeros everywhere else. Only 2D 

docs/Operators.md

@@ -33,6 +33,7 @@
   * <a href="#Dropout">Dropout</a>
   * <a href="#Elu">Elu</a>
   * <a href="#Equal">Equal</a>
+  * <a href="#Erf">Erf</a>
   * <a href="#Exp">Exp</a>
   * <a href="#Expand">Expand</a>
   * <a href="#EyeLike">EyeLike</a>
@@ -3252,6 +3253,57 @@ expect(node, inputs=[x, y], outputs=[z],
 </details>
 
 
+### <a name="Erf"></a><a name="erf">**Erf**</a>
+
+  Computes the error function of the given input tensor element-wise.
+
+#### Version
+
+This version of the operator has been available since version 9 of the default ONNX operator set.
+
+#### Inputs
+
+<dl>
+<dt><tt>input</tt> : T</dt>
+<dd>Input tensor</dd>
+</dl>
+
+#### Outputs
+
+<dl>
+<dt><tt>output</tt> : T</dt>
+<dd>The error function of the input tensor computed element-wise. It has the same shape and type of the input.</dd>
+</dl>
+
+#### Type Constraints
+
+<dl>
+<dt><tt>T</tt> : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)</dt>
+<dd>Constrain input and output types to all numeric tensors.</dd>
+</dl>
+
+
+#### Examples
+
+<details>
+<summary>erf</summary>
+
+```python
+node = onnx.helper.make_node(
+    'Erf',
+    inputs=['x'],
+    outputs=['y'],
+)
+
+x = np.random.randn(1, 3, 32, 32).astype(np.float32)
+y = np.vectorize(math.erf)(x).astype(np.float32)
+expect(node, inputs=[x], outputs=[y],
+       name='test_erf')
+```
+
+</details>
+
+
 ### <a name="Exp"></a><a name="exp">**Exp**</a>
 
   Calculates the exponential of the given input tensor, element-wise.

docs/TestCoverage.md

@@ -5,7 +5,7 @@
 * [Overall Test Coverage](#overall-test-coverage)
 # Node Test Coverage
 ## Summary
-Node tests have covered 104/111 (93.69%, 5 generators excluded) common operators.
+Node tests have covered 105/112 (93.75%, 5 generators excluded) common operators.
 
 Node tests have covered 2/12 (16.67%, 0 generators excluded) experimental operators.
 
@@ -1889,6 +1889,27 @@ expect(node, inputs=[x, y], outputs=[z],
 </details>
 
 
+### Erf
+There are 1 test cases, listed as following:
+<details>
+<summary>erf</summary>
+
+```python
+node = onnx.helper.make_node(
+    'Erf',
+    inputs=['x'],
+    outputs=['y'],
+)
+
+x = np.random.randn(1, 3, 32, 32).astype(np.float32)
+y = np.vectorize(math.erf)(x).astype(np.float32)
+expect(node, inputs=[x], outputs=[y],
+       name='test_erf')
+```
+
+</details>
+
+
 ### Exp
 There are 1 test cases, listed as following:
 <details>

onnx/backend/test/case/node/erf.py

@@ -0,0 +1,28 @@
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import math
+
+import numpy as np  # type: ignore
+
+import onnx
+from ..base import Base
+from . import expect
+
+
+class Erf(Base):
+
+    @staticmethod
+    def export():  # type: () -> None
+        node = onnx.helper.make_node(
+            'Erf',
+            inputs=['x'],
+            outputs=['y'],
+        )
+
+        x = np.random.randn(1, 3, 32, 32).astype(np.float32)
+        y = np.vectorize(math.erf)(x).astype(np.float32)
+        expect(node, inputs=[x], outputs=[y],
+               name='test_erf')

onnx/backend/test/data/node/test_erf/model.onnx

@@ -0,0 +1,15 @@
+backend-test:Q
+
+
x
y"Erftest_erfZ
+
x
+

+

+
+ 
+ b
+
y
+

+

+
+ 
+ B	

onnx/backend/test/data/node/test_erf/test_data_set_0/input_0.pb

@@ -1192,4 +1192,26 @@ ONNX_OPERATOR_SET_SCHEMA(
             OpSchema::all_numeric_types(),
             "Constrain input and output types to all numeric tensors.")
         .TypeAndShapeInferenceFunction(propagateShapeAndTypeFromFirstInput));
+
+static const char* Erf_ver9_doc = R"DOC(
+Computes the error function of the given input tensor element-wise.
+)DOC";
+
+ONNX_OPERATOR_SET_SCHEMA(
+	Erf,
+	9,
+	OpSchema()
+		.SetDoc(Erf_ver9_doc)
+		.Input(0, "input", "Input tensor", "T")
+		.Output(
+			0,
+			"output",
+			"The error function of the input tensor "
+			"computed element-wise. It has the same shape and type of the input.",
+			"T")
+		.TypeConstraint(
+			"T",
+            OpSchema::all_numeric_types(),
+            "Constrain input and output types to all numeric tensors.")
+        .TypeAndShapeInferenceFunction(propagateShapeAndTypeFromFirstInput));
 } // namespace ONNX_NAMESPACE

onnx/backend/test/data/node/test_erf/test_data_set_0/output_0.pb

@@ -490,6 +490,7 @@ class ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, Atanh);
 class ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, IsNaN);
 class ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, Sign);
 class ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, Scan);
+class ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, Erf);
 
 // Iterate over schema from ai.onnx version 9
 class OpSet_Onnx_ver9 {
@@ -520,6 +521,7 @@ class OpSet_Onnx_ver9 {
     fn(GetOpSchema<ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, IsNaN)>());
     fn(GetOpSchema<ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, Sign)>());
     fn(GetOpSchema<ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, Scan)>());
+    fn(GetOpSchema<ONNX_OPERATOR_SET_SCHEMA_CLASS_NAME(Onnx, 9, Erf)>());
   }
   static void ForEachFunctionBuilder(
       std::function<void(FunctionBuilder&&)> fn) {