Cyril Moineau · 564931a0 · 633a061f · 0a4aba96 · b5fa0429 · 2f2535bb
--- a/aidge_core/unit_tests/test_forwardDType.py 0 → 100644

+ 415

− 0
+++ b/aidge_core/unit_tests/test_forwardDType.py 0 → 100644

+ 415

− 0
+"""
+Copyright (c) 2023 CEA-List
+
+This program and the accompanying materials are made available under the
+terms of the Eclipse Public License 2.0 which is available at
+http://www.eclipse.org/legal/epl-2.0.
+
+SPDX-License-Identifier: EPL-2.0
+"""
+
+import unittest
+import aidge_core
+import numpy as np
+
+# List of all dtype defined by Aidge
+ALL_AIDGE_DTYPE = [i for i in aidge_core.dtype.__members__.values() if i != aidge_core.dtype.any]
+
+oh_no =[]
+
+class test_forwardDType(unittest.TestCase):
+    def setUp(self):
+        pass
+
+    def tearDown(self):
+        pass
+
+    ### HELPER FUNCTIONS ###
+    def verify_node_out_dtype(self, node, out_dtype):
+        """Helper function to verify output data type of a node
+        """
+        operator = node.get_operator()
+        self.assertEqual(operator.nb_outputs(), len(out_dtype), "Error in test design, the number of outputs provided does not correspond to the number of outputs of the operator.")
+        for out_idx in range(operator.nb_outputs()):
+            tensor_dtype = operator.get_output(out_idx).dtype()
+            self.assertEqual(tensor_dtype, out_dtype[out_idx], f"Node {node.name()}({node.type()}) output#{out_idx} is {tensor_dtype}, expected {out_dtype[out_idx]}")
+
+    def run_node_test(self, node, in_dtype, out_dtype):
+        """Run forwardDType unit test on the graph
+
+        :param graph: GraphView to call forwardDtype on
+        :type graph: aidge_core.GraphView
+        :param in_dtype: List of input type to forward
+        :type in_dtype: List[aidge_core.dtype]
+        :param out_dtype: List of expected output type
+        :type out_dtype: List[aidge_core.dtype]
+        """
+        op = node.get_operator()
+
+        for in_idx in range(len(in_dtype)):
+            in_tensor = aidge_core.Tensor()
+            in_tensor.set_datatype(in_dtype[in_idx])
+            op.set_input(in_idx, in_tensor)
+
+        self.assertTrue(op.forward_dtype(), "Forward data type failed")
+        self.verify_node_out_dtype(node, out_dtype)
+
+    def run_graph_test(self, graph, in_dtype, out_dtype):
+        """Run forwardDType unit test on the graph
+
+        :param graph: GraphView to call forwardDtype on
+        :type graph: aidge_core.GraphView
+        :param in_dtype: List of input type to forward
+        :type in_dtype: List[aidge_core.dtype]
+        :param out_dtype: Dictionary of node name and expected output type
+        :type out_dtype: Dict[str: List[aidge_core.dtype]]
+        """
+        # Loop to create an empty tensor for each operator outputs
+        # This replace a forwardDims!
+        # for node in graph.get_nodes():
+        #     op = node.get_operator()
+        #     if op.type() == aidge_core.ProducerOp.Type and op.attr.constant:
+        #         # Cannot set_output for constant Producer
+        #         continue
+        #     for out_idx in range(op.nb_outputs()):
+        #         out_tensor = aidge_core.Tensor()
+        #         oh_no.append(out_tensor)
+        #         op.set_output(out_idx, out_tensor)
+
+        self.assertTrue(graph.forward_dtype(in_dtype), "Forward data type failed")
+        for node in graph.get_nodes():
+            if node.name() not in out_dtype:
+                print(f"Warning: {node.name()}({node.type()}) if not tested!")
+            else:
+                self.verify_node_out_dtype(node, out_dtype[node.name()])
+
+    ### TESTING_OPERATORS ###
+    # Please ensure test cases are written in alphabetic order!
+
+    def test_Abs_forward_dtype(self):
+        pass
+
+    def test_Add_forward_dtype(self):
+        for in_dtype in ALL_AIDGE_DTYPE:
+            with self.subTest(dtype=f"Add forward_dtype: {in_dtype}"):
+                node = aidge_core.Div(name="add")
+                self.run_node_test(node, [in_dtype, in_dtype], [in_dtype])
+
+    def test_And_forward_dtype(self):
+        pass
+
+    def test_ArgMax_forward_dtype(self):
+        pass
+
+    def test_Atan_forward_dtype(self):
+        pass
+
+    def test_AvgPooling_forward_dtype(self):
+        pass
+
+    def test_BatchNorm_forward_dtype(self):
+        pass
+
+    def test_BitShift_forward_dtype(self):
+        pass
+
+    def test_Cast_forward_dtype(self):
+        for cast_dtype in ALL_AIDGE_DTYPE:
+            for in_dtype in ALL_AIDGE_DTYPE:
+                with self.subTest(dtype=f"Cast[{in_dtype}] forward_dtype:  {cast_dtype}"):
+                    cast = aidge_core.Cast(cast_dtype, name="Cast")
+                    # Whatever input type, expected out type is cast_dtype
+                    self.run_node_test(cast, [in_dtype], [cast_dtype])
+
+    def test_Clip_forward_dtype(self):
+        pass
+
+    def test_Concat_forward_dtype(self):
+        pass
+
+    def test_ConstantOfShape_forward_dtype(self):
+        pass
+
+    def test_Conv_forward_dtype(self):
+        pass
+
+    def test_ConvDepthWise_forward_dtype(self):
+        pass
+
+    def test_ConvTranspose_forward_dtype(self):
+        pass
+
+    def test_CryptoHash_forward_dtype(self):
+        pass
+
+    def test_DepthToSpace_forward_dtype(self):
+        pass
+
+    def test_Div_forward_dtype(self):
+        for in_dtype in ALL_AIDGE_DTYPE:
+            with self.subTest(dtype=f"Div forward_dtype: {in_dtype}"):
+                node = aidge_core.Div(name="Div")
+                self.run_node_test(node, [in_dtype, in_dtype], [in_dtype])
+
+    def test_Equal_forward_dtype(self):
+        pass
+
+    def test_Erf_forward_dtype(self):
+        pass
+
+    def test_Expand_forward_dtype(self):
+        pass
+
+    def test_FC_forward_dtype(self):
+        test_cases = [
+            ("float32", [aidge_core.dtype.float32] * 3, [aidge_core.dtype.float32]),
+            ("int8", [aidge_core.dtype.int8, aidge_core.dtype.int8, aidge_core.dtype.int32], [aidge_core.dtype.int32]),
+        ]
+
+        for name, in_dtype, out_dtype in test_cases:
+            with self.subTest(dtype=name):
+                node = aidge_core.FC(1, 1, name="FC")
+                self.run_node_test(node, in_dtype, out_dtype)
+
+    def test_Flatten_forward_dtype(self):
+        pass
+
+    def test_Fold_forward_dtype(self):
+        pass
+
+    def test_Gather_forward_dtype(self):
+        pass
+
+    def test_GenericOperator_forward_dtype(self):
+        pass
+
+    def test_GlobalAveragePooling_forward_dtype(self):
+        pass
+
+    def test_GridSample_forward_dtype(self):
+        pass
+
+    def test_Heaviside_forward_dtype(self):
+        pass
+
+    def test_ILayerNorm_forward_dtype(self):
+        pass
+
+    def test_Identity_forward_dtype(self):
+        pass
+
+    def test_LRN_forward_dtype(self):
+        pass
+
+    def test_LeakyReLU_forward_dtype(self):
+        pass
+
+    def test_Ln_forward_dtype(self):
+        pass
+
+    def test_MatMul_forward_dtype(self):
+        pass
+
+    def test_MaxPooling_forward_dtype(self):
+        pass
+
+    def test_Memorize_forward_dtype(self):
+        pass
+
+    def test_MetaOperator_forward_dtype(self):
+        pass
+
+    def test_MetaOperatorDefs_forward_dtype(self):
+        pass
+
+    def test_Mod_forward_dtype(self):
+        pass
+
+    def test_Move_forward_dtype(self):
+        pass
+
+    def test_Mul_forward_dtype(self):
+        for in_dtype in ALL_AIDGE_DTYPE:
+            with self.subTest(dtype=f"Mul forward_dtype: {in_dtype}"):
+                node = aidge_core.Mul(name="Mul")
+                self.run_node_test(node, [in_dtype, in_dtype], [in_dtype])
+
+    def test_Pad_forward_dtype(self):
+        pass
+
+    def test_Pop_forward_dtype(self):
+        pass
+
+    def test_Pow_forward_dtype(self):
+        pass
+
+    def test_Producer_forward_dtype(self):
+        pass
+
+    def test_ReLU_forward_dtype(self):
+        for in_dtype in ALL_AIDGE_DTYPE:
+            with self.subTest(dtype=f"ReLU forward_dtype: {in_dtype}"):
+                node = aidge_core.ReLU(name="Relu")
+                self.run_node_test(node, [in_dtype], [in_dtype])
+
+    def test_ReduceMean_forward_dtype(self):
+        pass
+
+    def test_ReduceSum_forward_dtype(self):
+        pass
+
+    def test_Reshape_forward_dtype(self):
+        pass
+
+    def test_Resize_forward_dtype(self):
+        pass
+
+    def test_Round_forward_dtype(self):
+        pass
+
+    def test_Scaling_forward_dtype(self):
+        pass
+
+    def test_Select_forward_dtype(self):
+        pass
+
+    def test_Shape_forward_dtype(self):
+        pass
+
+    def test_ShiftGELU_forward_dtype(self):
+        pass
+
+    def test_ShiftMax_forward_dtype(self):
+        pass
+
+    def test_Sigmoid_forward_dtype(self):
+        pass
+
+    def test_Slice_forward_dtype(self):
+        pass
+
+    def test_Softmax_forward_dtype(self):
+        pass
+
+    def test_Split_forward_dtype(self):
+        pass
+
+    def test_Sqrt_forward_dtype(self):
+        pass
+
+    def test_Squeeze_forward_dtype(self):
+        pass
+
+    def test_Stack_forward_dtype(self):
+        pass
+
+    def test_Sub_forward_dtype(self):
+        for in_dtype in ALL_AIDGE_DTYPE:
+            with self.subTest(dtype=f"Add forward_dtype: {in_dtype}"):
+                node = aidge_core.Sub(name="sub")
+                self.run_node_test(node, [in_dtype, in_dtype], [in_dtype])
+
+    def test_Tanh_forward_dtype(self):
+        pass
+
+    def test_Transpose_forward_dtype(self):
+        pass
+
+    def test_Unfold_forward_dtype(self):
+        pass
+
+    def test_Unsqueeze_forward_dtype(self):
+        pass
+
+    def test_WeightInterleaving_forward_dtype(self):
+        pass
+
+
+    ### TESTING GRAPH ###
+
+    def test_shuffle_net(self):
+        # Declaring constant values
+        prod_two_a = aidge_core.Producer(aidge_core.Tensor(np.array(2, dtype=np.int64)), "two_a", constant=True)
+        prod_two_b = aidge_core.Producer(aidge_core.Tensor(np.array(2, dtype=np.int64)), "two_b", constant=True)
+
+        # Declaring operators
+        shape_op_1     = aidge_core.Shape(name="shape_op_1")
+        shape_op_2     = aidge_core.Shape(name="shape_op_2")
+        shape_op_3     = aidge_core.Shape(name="shape_op_3")
+        shape_op_4     = aidge_core.Shape(name="shape_op_4")
+        gather_op_1    = aidge_core.Gather(axis = 0, indices = [0], name="gather_op_1")
+        gather_op_2    = aidge_core.Gather(axis = 0, indices = [1], name="gather_op_2")
+        gather_op_3    = aidge_core.Gather(axis = 0, indices = [2], name="gather_op_3")
+        gather_op_4    = aidge_core.Gather(axis = 0, indices = [3], name="gather_op_4")
+        div_op         = aidge_core.Div(name="div_op")
+
+        u_op_1         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_1")
+        u_op_2         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_2")
+        u_op_3         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_3")
+        u_op_4         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_4")
+        u_op_5         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_5")
+        u_op_6         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_6")
+        u_op_7         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_7")
+        u_op_8         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_8")
+        u_op_9         = aidge_core.Unsqueeze(axes = [0], name="unsqueeze_op_9")
+        concat_op_1    = aidge_core.Concat(5, name="concat_op_1")
+        concat_op_2    = aidge_core.Concat(4, name="concat_op_2")
+        reshape_op_1   = aidge_core.Reshape(name=  "reshape_op_1")
+        reshape_op_2   = aidge_core.Reshape(name=  "reshape_op_2")
+        transpose_op_1 = aidge_core.Transpose([0, 2, 1, 3, 4], name="transpose_op_1")
+
+        # Declaring Connectors
+        x = aidge_core.Connector(aidge_core.Identity(f"Input"))
+        a = aidge_core.Connector(prod_two_a)
+        b = aidge_core.Connector(prod_two_b)
+
+        # Graph creation using functional declaration
+        x1 = shape_op_1(x)
+        x2 = shape_op_2(x)
+        x3 = shape_op_3(x)
+        x4 = shape_op_4(x)
+        n = gather_op_1(x1)
+        c = gather_op_2(x2)
+        h = gather_op_3(x3)
+        w = gather_op_4(x4)
+
+        shape_1 = concat_op_1(u_op_1(n), u_op_2(a), u_op_3(div_op(c, b)), u_op_4(h), u_op_5(w))
+        shape_2 = concat_op_2(u_op_6(n), u_op_7(c), u_op_8(h), u_op_9(w))
+
+        y = reshape_op_2(transpose_op_1(reshape_op_1(x, shape_1)), shape_2)
+
+        shuffle_net_graph = aidge_core.generate_graph([y])
+        for in_dtype in ALL_AIDGE_DTYPE:
+            with self.subTest(dtype=f"ShuffleNet {in_dtype}"):
+                output_dtype = {
+                    "shape_op_1":     [aidge_core.dtype.int64],
+                    "shape_op_2":     [aidge_core.dtype.int64],
+                    "shape_op_3":     [aidge_core.dtype.int64],
+                    "shape_op_4":     [aidge_core.dtype.int64],
+                    "gather_op_1":    [aidge_core.dtype.int64],
+                    "gather_op_3":    [aidge_core.dtype.int64],
+                    "gather_op_2":    [aidge_core.dtype.int64],
+                    "gather_op_4":    [aidge_core.dtype.int64],
+                    "div_op":         [aidge_core.dtype.int64],
+                    "unsqueeze_op_1": [aidge_core.dtype.int64],
+                    "unsqueeze_op_2": [aidge_core.dtype.int64],
+                    "unsqueeze_op_3": [aidge_core.dtype.int64],
+                    "unsqueeze_op_4": [aidge_core.dtype.int64],
+                    "unsqueeze_op_5": [aidge_core.dtype.int64],
+                    "unsqueeze_op_6": [aidge_core.dtype.int64],
+                    "unsqueeze_op_7": [aidge_core.dtype.int64],
+                    "unsqueeze_op_8": [aidge_core.dtype.int64],
+                    "unsqueeze_op_9": [aidge_core.dtype.int64],
+                    "concat_op_1":    [aidge_core.dtype.int64],
+                    "concat_op_2":    [aidge_core.dtype.int64],
+                    "two_a":          [aidge_core.dtype.int64],
+                    "two_b":          [aidge_core.dtype.int64],
+                    "reshape_op_1":   [in_dtype],
+                    "reshape_op_2":   [in_dtype],
+                    "transpose_op_1": [in_dtype],
+                    "Input":          [in_dtype]
+                }
+                self.run_graph_test(shuffle_net_graph, [in_dtype], output_dtype)
+
+if __name__ == '__main__':
+    unittest.main()