[NNVM] Support argmax/argmin in tensorflow frontend (#1514)

Author: sergei-mironov

nnvm/python/nnvm/frontend/tensorflow.py

@@ -91,6 +91,20 @@ def _impl(inputs, attr, *args):
         return AttrCvt(op_name="__pow_scalar__", extras={'scalar': -0.5})(inputs, attr)
     return _impl
 
+def _argx(func, func_name):
+    """ A common wrapper for argmin and argmax operations """
+    def _impl(inputs, attr, params):
+        try:
+            # In Tensorflow, `axis` argument is a Tensor, not attribute. We
+            # support the case where it inputs from a scalar constant.
+            axis_input_name = inputs[1].list_output_names()[0]
+            axis_input_vlaue = params[axis_input_name].asnumpy()[0]
+        except (IndexError, KeyError):
+            raise TypeError( \
+                "Unsupported argument for `{}` : `axis` should be a constant".format(func_name))
+        return func(inputs[0], axis=axis_input_vlaue, keepdims=False)
+    return _impl
+
 def _elemwise(name):
     def _impl(inputs, attr, *args):
         assert len(inputs) == 2, "Math op take 2 inputs, {} given".format(len(inputs))
@@ -664,6 +678,8 @@ def _impl(inputs, in_state_c, in_state_h, attr, params):
 # for 1 to N mapping(composed), use custom callable functions
 # for N to 1 mapping, currently not supported(?)
 _convert_map = {
+    'ArgMax'                            : _argx(_sym.argmax, 'argmax'),
+    'ArgMin'                            : _argx(_sym.argmin, 'argmin'),
     'AvgPool'                           : _pooling('avg_pool'),
     'BatchNormWithGlobalNormalization'  : _batch_norm(),
     'BiasAdd'                           : _bias_add(),
@@ -879,6 +895,28 @@ def _get_abs_layer_name(node):
                                                      params, num_layers)
         return sym
 
+
+def _parse_import_prerequisites(graph):
+    """ Calculate the named preconditions from TensorFlow `graph`.
+    Return prerequisites for parsing:
+     a. Set of operator names which don't have their mapping in TVM, i.e.
+        which are not supported
+    """
+    missing_operators = set()
+    for node in graph.node:
+        if node.op == "Placeholder":
+            pass
+        elif node.op == "Const":
+            pass
+        else:
+            if any([node.op in t for t in [_identity_list, _convert_map, _convert_map_rnn]]):
+                pass
+            else:
+                missing_operators.add(node.op)
+
+    return missing_operators
+
+
 class GraphProto(object):
     """ A helper class for handling nnvm graph copying from Tensorflow GraphDef.
     Definition:
@@ -901,7 +939,7 @@ def from_tensorflow(self, graph):
         Follow the tensorflow graph definition to parse and convert it to NNVM.
         Some of the assumptions listed below.
 
-            -> First Const or Placeholder node will be considered as graph input.
+            -> First Placeholder or Const node will be considered as graph input.
             -> Rest all Const nodes are params.
             -> Last node is assumed as graph output.
             -> _output_shapes : Attribute should present in the tenserflow forzen graph.
@@ -910,6 +948,7 @@ def from_tensorflow(self, graph):
             -> CheckNumerics: No implementation as of now for this.
                               Just copies input to output.
 
+        TODO: Change algorithm to stop treating first 'Const' in a special way.
 
         Parameters
         ----------
@@ -923,23 +962,25 @@ def from_tensorflow(self, graph):
         params : dict
             A dict of name: tvm.nd.array pairs, used as pretrained weights
         """
-        # Parse throught all nodes and start extracting
-        # params aka Const nodes
-        # input nodes  : First const node
-        # normal nodes : other normal nodes
 
         try:
             from tensorflow.python.framework import tensor_util
         except ImportError as e:
             raise ImportError(
                 "Unable to import tensorflow which is required {}".format(e))
 
+        missing_operators = _parse_import_prerequisites(graph)
+
+        if missing_operators:
+            raise NotImplementedError( \
+                "The following operators are not implemented: {}".format(missing_operators))
+
+        # Parse the nodes to re-create TF graph using Symbol API of NNVM
         for node in graph.node:
             # Tensorflow doesn't have seperate list for params extraction.
             # Operator name 'Const' is treated as a parameter to build NNVM params dict.
             input_shapes = {}
             if node.op == "Placeholder":
-                # Assuming only one input graph with type 'Placeholder'
                 self._input_node = node.name
                 self._num_input += 1
 
@@ -954,7 +995,6 @@ def from_tensorflow(self, graph):
                     raise NotImplementedError( \
                         "Please freeze the graph with add_shapes=True")
             elif node.op == "Const":
-                # Assuming first Const node as Graph Input node
                 if self._input_node == '':
                     self._input_node = node.name
                     self._num_input += 1
@@ -997,7 +1037,7 @@ def from_tensorflow(self, graph):
                 # Pass the node name too in attr
                 attr["_node_name"] = node.name
 
-                #ToDo: Some of the tensorflow operators maintain internaly maintain
+                #ToDo: Some of the tensorflow operators internaly maintain
                 #execution layers and its output name will the layer number along with
                 #graph node name.eg: Node name:- 'Model/RNN/cell_0/RnnCell', but the
                 #output name will be 'Model/RNN/cell_0/RnnCell:0'. In this case,

nnvm/tests/python/frontend/tensorflow/test_forward.py

@@ -404,6 +404,37 @@ def test_forward_sigmoid():
 
     _test_sigmoid(np.random.uniform(size=(3, 4, 4, 3)).astype('float32'))
 
+#######################################################################
+# Argmin/Argmax
+# -------------
+
+def _test_argx(func, data, **kwargs):
+
+    with tf.Graph().as_default():
+        inp = constant_op.constant(data, shape=data.shape, dtype=data.dtype, name="c0")
+
+        # pylint: disable=unused-variable
+        out = func(inp, name="argx0", **kwargs)
+        # pylint: enable=unused-variable
+
+        with tf.Session() as sess:
+            graph_def = tf.graph_util.convert_variables_to_constants(
+                sess=sess,
+                input_graph_def=sess.graph.as_graph_def(add_shapes=True),
+                output_node_names=["argx0"])
+
+            tf_output = run_tf_graph(sess, data, input_node="c0:0", output_node="argx0:0")
+            tvm_output = run_tvm_graph(graph_def, data, "c0", tf_output.shape, output_dtype='int32')
+
+            np.testing.assert_allclose(tf_output, tvm_output, atol=1e-5, rtol=1e-5)
+
+            sess.close()
+
+def test_argmin_argmax():
+    for axis in [None,0,1,2]:
+        data = np.random.uniform(size=(8,4,9)).astype('float32')
+        _test_argx(tf.argmax, data=data, axis=axis)
+        _test_argx(tf.argmin, data=data, axis=axis)
 
 #######################################################################
 # Variable