Revert "perf: Prefer generator expressions over list comprehensions (#2486)" (#2491)

This reverts commit 37b9aeb.

Author: DawerG

coremltools/converters/_converters_entry.py

@@ -750,14 +750,14 @@ def _validate_outputs_argument(outputs):
                                  'or of types ct.ImageType/ct.TensorType'
         if isinstance(outputs[0], str):
             # if one of the elements is a string, all elements must be strings
-            if not all(isinstance(t, str) for t in outputs):
+            if not all([isinstance(t, str) for t in outputs]):
                 raise ValueError(msg_inconsistent_types)
             return outputs, [TensorType(name=name) for name in outputs]
 
         if isinstance(outputs[0], InputType):
-            if not all(isinstance(t, TensorType) or isinstance(t, ImageType) for t in outputs):
+            if not all([isinstance(t, TensorType) or isinstance(t, ImageType) for t in outputs]):
                 raise ValueError(msg_inconsistent_types)
-            if any(t.shape is not None for t in outputs):
+            if any([t.shape is not None for t in outputs]):
                 msg = "The 'shape' argument must not be specified for the outputs, since it is " \
                       "automatically inferred from the input shapes and the ops in the model"
                 raise ValueError(msg)
@@ -777,9 +777,9 @@ def _validate_outputs_argument(outputs):
             output_names = [t.name for t in outputs]
             # verify that either all of the entries in output_names is "None" or none of them is "None"
             msg_consistent_names = 'Either none or all the outputs must have the "name" argument specified'
-            if output_names[0] is None and not all(name is None for name in output_names):
+            if output_names[0] is None and not all([name is None for name in output_names]):
                 raise ValueError(msg_consistent_names)
-            if output_names[0] is not None and not all(name is not None for name in output_names):
+            if output_names[0] is not None and not all([name is not None for name in output_names]):
                 raise ValueError(msg_consistent_names)
             if output_names[0] is not None:
                 if len(set(output_names)) != len(output_names):
@@ -914,7 +914,7 @@ def _flatten_list(_inputs):
         if inputs is not None:
             raise_if_duplicated(inputs)
 
-        if inputs is not None and not all(isinstance(_input, InputType) for _input in inputs):
+        if inputs is not None and not all([isinstance(_input, InputType) for _input in inputs]):
             raise ValueError("Input should be a list of TensorType or ImageType")
 
     elif exact_source == "pytorch":

coremltools/converters/mil/backend/mil/load.py

@@ -551,7 +551,7 @@ def remove_output(block, prob_var):
         classes = classes.splitlines()
     elif isinstance(classes_in, list):  # list[int or str]
         classes = classes_in
-        assert all(isinstance(x, (int, str)) for x in classes), message
+        assert all([isinstance(x, (int, str)) for x in classes]), message
     else:
         raise ValueError(message)
 

coremltools/converters/mil/backend/nn/load.py

@@ -266,7 +266,7 @@ def load(prog, **kwargs):
 
     proto = builder.spec
     # image input
-    has_image_input = any(isinstance(s, ImageType) for s in input_types)
+    has_image_input = any([isinstance(s, ImageType) for s in input_types])
     if has_image_input:
         proto = _convert_to_image_input(proto, input_types,
                                         skip_model_load=kwargs.get("skip_model_load", False))
@@ -284,7 +284,7 @@ def load(prog, **kwargs):
                 shape = var.sym_type.get_shape()
                 if any_variadic(shape):
                     raise ValueError("Variable rank model outputs, that are ImageTypes, are not supported")
-                if any(is_symbolic(d) for d in shape):
+                if any([is_symbolic(d) for d in shape]):
                     raise NotImplementedError("Image output '{}' has symbolic dimensions in its shape".
                                               format(var.name))
                 _validate_image_input_output_shapes(output_types[i].color_layout, shape, var.name, is_input=False)

coremltools/converters/mil/backend/nn/op_mapping.py

@@ -196,7 +196,7 @@ def _try_convert_global_pool(const_context, builder, op, mode):
 
     if tuple(op.outputs[0].shape[:-2]) != tuple(op.inputs["x"].shape[:-2]):
         return False
-    if not all(s == 1 for s in op.outputs[0].shape[-2:]):
+    if not all([s == 1 for s in op.outputs[0].shape[-2:]]):
         return False
 
     builder.add_pooling(
@@ -795,11 +795,11 @@ def _add_elementwise_binary(
                 # INTERNAL_MUL_XYKN not implemented
                 continue
             if all(shape_x[indices] == shape_y[indices]):
-                if all(True if i in indices else s == 1 for i, s in enumerate(shape_x)):
+                if all([True if i in indices else s == 1 for i, s in enumerate(shape_x)]):
                     internal_y = op.x
                     internal_x = op.y
                     break
-                if all(True if i in indices else s == 1 for i, s in enumerate(shape_y)):
+                if all([True if i in indices else s == 1 for i, s in enumerate(shape_y)]):
                     internal_x = op.x
                     internal_y = op.y
                     break
@@ -3323,7 +3323,7 @@ def stack(const_context, builder, op):
 def split(const_context, builder, op):
     split = op.sizes
     split = [size for size in split if size != 0]
-    has_equal_splits = all(size == split[0] for size in split)
+    has_equal_splits = all([size == split[0] for size in split])
     num_splits = len(split)
     output_names = [op.outputs[i].name for i in range(len(op.sizes)) if op.sizes[i] != 0]
 
@@ -3545,7 +3545,7 @@ def _realloc_list(const_context, builder, ls_var, index_var, value_var, mode):
 
     # check if elem_shape is runtime-determined
     elem_shape = tuple(value_var.shape)
-    has_dynamic_shape = any(is_symbolic(i) for i in elem_shape)
+    has_dynamic_shape = any([is_symbolic(i) for i in elem_shape])
 
     # get the fill shape of the tensor array
     # [length, elem_dim1, elem_dim2, ...]

coremltools/converters/mil/converter.py

@@ -58,7 +58,7 @@ def __call__(self, model, *args, **kwargs):
                         type(inputs)
                     )
                 )
-            if not all(isinstance(i, input_types.InputType) for i in inputs):
+            if not all([isinstance(i, input_types.InputType) for i in inputs]):
                 raise ValueError(
                     "Type of inputs should be list or tuple of TensorType or ImageType, got {} instead.".format(
                         [type(i) for i in inputs]

coremltools/converters/mil/debugging_utils.py

@@ -84,7 +84,7 @@ def validate_inputs(func, input_vars):
                 else:
                     input_values.append(v)
 
-            if all(x in reachable_vars for x in input_values):
+            if all([x in reachable_vars for x in input_values]):
                 reachable_vars.update(op.outputs)
 
         for out in func.outputs:

coremltools/converters/mil/frontend/_utils.py

@@ -293,7 +293,7 @@ def _swap(a, b):
         return b, a
 
     a_var, b_var = vars
-    is_dynamic = any(any_symbolic(var.shape) for var in vars)
+    is_dynamic = any([any_symbolic(var.shape) for var in vars])
     # list of equations supported for explicit mil translations
     vec_bnqd_bnkd_bnqk = (
         [0, 1, 2, 3],
@@ -436,10 +436,10 @@ def get_output_names(outputs) -> Optional[List[str]]:
 
     output_names = None
     if outputs is not None:
-        assert all(isinstance(t, InputType) for t in outputs), \
+        assert all([isinstance(t, InputType) for t in outputs]), \
             "outputs must be a list of ct.ImageType or ct.TensorType"
         output_names = [t.name for t in outputs]
-        if all(name is None for name in output_names):
+        if all([name is None for name in output_names]):
             output_names = None
     return output_names
 

coremltools/converters/mil/frontend/tensorflow/converter.py

@@ -179,7 +179,7 @@ def __init__(
                         type(inputs)
                     )
                 )
-            if not all(isinstance(i, InputType) for i in inputs):
+            if not all([isinstance(i, InputType) for i in inputs]):
                 raise ValueError(
                     "Type of inputs should be list or tuple of TensorType or ImageType, got {} instead.".format(
                         [type(i) for i in inputs]
@@ -238,7 +238,7 @@ def __init__(
         for inputtype in self.inputs:
             if not isinstance(inputtype.shape, InputShape):
                 continue
-            if any(isinstance(s, RangeDim) for s in inputtype.shape.shape):
+            if any([isinstance(s, RangeDim) for s in inputtype.shape.shape]):
                 continue
             if inputtype.name not in graph:
                 raise ValueError(

coremltools/converters/mil/frontend/tensorflow/ops.py

@@ -633,7 +633,7 @@ def ExtractImagePatches(context, node):
     padding = node.attr.get("padding")
     if x.rank != 4:
         raise ValueError("input for ExtractImagePatches should be a 4D tensor.")
-    if not all(rate == 1 for rate in rates):
+    if not all([rate == 1 for rate in rates]):
         raise NotImplementedError(
             "only rates with all 1s is implemented for ExtractImagePatches."
         )
@@ -3022,7 +3022,7 @@ def Pack(context, node):
         else:
             x = mb.expand_dims(x=values[0], axes=[axis], name=node.name)
     else:
-        if all(_is_scalar(input.sym_type) for input in values):
+        if all([_is_scalar(input.sym_type) for input in values]):
             x = mb.concat(values=values, axis=axis, name=node.name)
         else:
             x = mb.stack(values=values, axis=axis, name=node.name)

coremltools/converters/mil/frontend/tensorflow/ssa_passes/tf_lstm_to_core_lstm.py

@@ -106,7 +106,7 @@ def _try_get_last_cell_state_in_tf_lstm_block(op: Operation) -> Var:
         return cs
     if len(cs.consuming_blocks) > 1:
         return None
-    if not all(child_op.op_type == "slice_by_index" for child_op in cs.child_ops):
+    if not all([child_op.op_type == "slice_by_index" for child_op in cs.child_ops]):
         return None
     child_ops = cs.child_ops[:]
     block = op.enclosing_block

coremltools/converters/mil/frontend/tensorflow/test/test_ops.py

@@ -1767,7 +1767,7 @@ def build_model_static_weights(x):
             )
 
         test_static_W()
-        if not any(True if d > 1 else False for d in dilations):
+        if not any([True if d > 1 else False for d in dilations]):
             if backend[0] == "neuralnetwork":
                 pytest.skip("dynamic conv with groups > 1 is not supported on the neuralnetwork backend")
             test_dynamic_W()

coremltools/converters/mil/frontend/tensorflow/test/testing_utils.py

@@ -164,7 +164,7 @@ def tf_graph_to_mlmodel(
             input_types.append(
                 ct.TensorType(name=input_placeholder.name.split(":")[0], shape=input_shape)
             )
-            if any(dim.value is None for dim in input_placeholder.shape):
+            if any([dim.value is None for dim in input_placeholder.shape]):
                 has_dynamic_shape = True
         if has_dynamic_shape:
             inputs_for_conversion = input_types

coremltools/converters/mil/frontend/torch/ops.py

@@ -6217,7 +6217,7 @@ def _check_args(tensor_inputs, indexing) -> None:
         assert isinstance(tensor_inputs, (list, tuple))
         if len(tensor_inputs) < 2:
             raise ValueError("Requires >= 2 tensor inputs.")
-        if any(tensor_input.rank > 1 for tensor_input in tensor_inputs):
+        if any([tensor_input.rank > 1 for tensor_input in tensor_inputs]):
             raise ValueError("meshgrid received non-1d tensor.")
 
         if indexing not in ("ij", "xy"):
@@ -7200,7 +7200,7 @@ def where(context, node):
     if not types.is_bool(cond.dtype):
         # cond must be bool type
         cond = mb.cast(x=cond, dtype="bool")
-    if not any(any_symbolic(x.shape) for x in (cond, a, b)):
+    if not any([any_symbolic(x.shape) for x in (cond, a, b)]):
         # broadcast all tensors to the same shape
         cond, a, b = _utils.pymil_broadcast_tensors([cond, a, b])
     result = mb.select(cond=cond, a=a, b=b, name=node.name)
@@ -7692,7 +7692,7 @@ def _pad_packed_sequence(context, node):
 
     # we only support pack and unpack translation for static tensor shape,
     # i.e., the three dimensions are all known during compile time.
-    if any(is_symbolic(x) for x in input_tensor.shape):
+    if any([is_symbolic(x) for x in input_tensor.shape]):
         raise NotImplementedError("Only static shape of PackedSequence object is supported.")
 
     # the input always has batch first layout.

coremltools/converters/mil/mil/block.py

@@ -1126,7 +1126,7 @@ def set_output_types(self, outputs: Optional[List["_input_types.InputType"]] = N
         if outputs is not None:
             if not (
                 isinstance(outputs, list)
-                and all(isinstance(out, _input_types.InputType) for out in outputs)
+                and all([isinstance(out, _input_types.InputType) for out in outputs])
             ):
                 raise TypeError(
                     "main outputs should be a list of type ct.TensorType or ct.ImageType"
@@ -1136,6 +1136,6 @@ def set_output_types(self, outputs: Optional[List["_input_types.InputType"]] = N
     def set_input_types(self, input_types: List["_input_types.InputType"]):
         if not isinstance(input_types, tuple):
             raise ValueError("main inputs should be tuple of TensorType or ImageType")
-        elif not all(isinstance(inp, _input_types.InputType) for inp in input_types):
+        elif not all([isinstance(inp, _input_types.InputType) for inp in input_types]):
             raise ValueError("main inputs should be tuple of InputSpec")
         self.input_types = input_types

coremltools/converters/mil/mil/ops/defs/iOS15/conv.py

@@ -182,7 +182,7 @@ def type_inference(self):
         custom_pad = None if self.pad_type.val != 'custom' else self.pad.val
 
         is_weight_dynamic = not self.weight.is_descendant_of_const
-        if is_weight_dynamic and any(True if d > 1 else False for d in dilations):
+        if is_weight_dynamic and any([True if d > 1 else False for d in dilations]):
             raise ValueError("Convolution with dynamic weights does not support dilations!")
 
         N = inshape[0]

coremltools/converters/mil/mil/ops/defs/iOS15/elementwise_binary.py

@@ -65,7 +65,7 @@ def _cast_check_value_inferene(self, a, b):
         """
         If one of the input is tensor, cast the result to tensor.
         """
-        to_cast = any(isinstance(x, np.ndarray) for x in (a, b))
+        to_cast = any([isinstance(x, np.ndarray) for x in [a, b]])
         result = self.get_operator()(a, b)
         return result if not to_cast else np.array(result)
 

coremltools/converters/mil/mil/ops/defs/iOS15/normalization.py

@@ -283,7 +283,7 @@ def type_inference(self):
 
         # check valid axes
         positive_axes = [axis + rank if axis < 0 else axis for axis in self.axes.val]
-        if not all(axis >= 0 and axis < rank for axis in positive_axes):
+        if not all([axis >= 0 and axis < rank for axis in positive_axes]):
             raise ValueError("axes must in the range of [-x.rank, x.rank-1].")
 
         # check shape of gamma and beta

coremltools/converters/mil/mil/ops/defs/iOS15/tensor_operation.py

@@ -1091,7 +1091,7 @@ def value_inference(self):
             symbolic_tensor = np.reshape(np.array(symbolic_tensor), shape)
             values.append(symbolic_tensor)
 
-        if any(val is None for val in values):
+        if any([val is None for val in values]):
             return None
 
         if not isinstance(values[0], np.ndarray) or values[0].shape == ():
@@ -1301,13 +1301,13 @@ def type_inference(self):
     @precondition(allow=VALUE | SYMBOL | NONE)
     def value_inference(self):
 
-        is_all_rank_zero = all(v.rank == 0 for v in self.values)
+        is_all_rank_zero = all([v.rank == 0 for v in self.values])
         values = [
             v.sym_val if v.sym_val is not None else get_new_symbol()
             for v in self.values
         ]
 
-        if any(is_symbolic(v) for v in values) and not is_all_rank_zero:
+        if any([is_symbolic(v) for v in values]) and not is_all_rank_zero:
             return None
 
         return np.stack(values, self.axis.val)

coremltools/converters/mil/mil/ops/tests/iOS14/test_scatter_gather.py

@@ -478,7 +478,7 @@ def prog(x):
             res = mb.gather(x=params, indices=indices, axis=-1)
             return res
 
-        if any(idx > 2 for idx in indices_val):
+        if any([idx > 2 for idx in indices_val]):
             with pytest.raises(IndexError, match="index 3 is out of bounds for axis 1 with size 3"):
                 mb.program(
                     input_specs=[mb.TensorSpec(shape=(1,), dtype=types.fp32)],
@@ -672,7 +672,7 @@ def prog(x):
             res = mb.gather_along_axis(x=params, indices=indices, axis=0)
             return res
 
-        if any(idx > 1 for sub_indices in indices_val for idx in sub_indices):
+        if any([idx > 1 for sub_indices in indices_val for idx in sub_indices]):
             with pytest.raises(IndexError, match="index 2 is out of bounds for axis 0 with size 2"):
                 mb.program(
                     input_specs=[mb.TensorSpec(shape=(1,), dtype=types.fp32)],

coremltools/converters/mil/mil/ops/tests/iOS17/test_scatter_gather.py

@@ -131,7 +131,7 @@ def build_dynamic(data, indices, updates):
 
         if not isinstance(expected_error_msg, tuple):
             expected_error_msg = expected_error_msg
-        assert any(err in str(excinfo.value) for err in expected_error_msg)
+        assert any([err in str(excinfo.value) for err in expected_error_msg])
 
 class TestScatterAlongAxis:
     @pytest.mark.parametrize(
@@ -224,7 +224,7 @@ def build_dynamic(data, indices, updates):
 
         if not isinstance(expected_error_msg, tuple):
             expected_error_msg = expected_error_msg
-        assert any(err in str(excinfo.value) for err in expected_error_msg)
+        assert any([err in str(excinfo.value) for err in expected_error_msg])
 
 
 class TestScatterNd:
@@ -295,7 +295,7 @@ def build_dynamic(data, indices, updates):
             )
         if not isinstance(expected_error_msg, tuple):
             expected_error_msg = expected_error_msg
-        assert any(err in str(excinfo.value) for err in expected_error_msg)
+        assert any([err in str(excinfo.value) for err in expected_error_msg])
 
 
 class TestGather(_TestGatherIOS16):
@@ -333,7 +333,7 @@ def prog(x):
                     input_specs=[mb.TensorSpec(shape=(1,), dtype=types.fp32)],
                     opset_version=backend.opset_version,
                 )(prog)
-        elif any(idx > 2 for idx in indices_val):
+        elif any([idx > 2 for idx in indices_val]):
             # If the indices are not validated during type inference for IOS17, the `gather` op's
             # value inference will raise error for out-of-bound index.
             with pytest.raises(IndexError, match="index 3 is out of bounds for axis 1 with size 3"):
@@ -410,7 +410,7 @@ def prog(x):
                     input_specs=[mb.TensorSpec(shape=(1,), dtype=types.fp32)],
                     opset_version=backend.opset_version,
                 )(prog)
-        elif any(idx > 1 for sub_indices in indices_val for idx in sub_indices):
+        elif any([idx > 1 for sub_indices in indices_val for idx in sub_indices]):
             # If the indices are not validated during type inference for IOS17, the `gather` op's
             # value inference will raise error for out-of-bound index.
             with pytest.raises(IndexError, match="index 2 is out of bounds for axis 0 with size 2"):

coremltools/converters/mil/mil/passes/defs/cleanup/noop_elimination.py

@@ -127,7 +127,7 @@ def remove_slice_by_index(op):
 
             if op.stride is not None and op.stride.val is not None:
                 stride = op.stride.val.flatten().tolist()
-                if any(x < 0 for x in stride):
+                if any([x < 0 for x in stride]):
                     return False
 
             if op.enclosing_block.try_replace_uses_of_var_after_op(

coremltools/converters/mil/mil/passes/defs/optimize_activation_quantization.py

@@ -161,7 +161,7 @@ def _try_match_and_transform_pattern(
 
         # Reject if 1st operation is not `conv`/`add`/`pool`.
         SUPPORTED_OP_TYPES = ["conv", "add", "avg_pool", "max_pool"]
-        if any(_check_child_op_type(dequantize_op, val) for val in SUPPORTED_OP_TYPES):
+        if any([_check_child_op_type(dequantize_op, val) for val in SUPPORTED_OP_TYPES]):
             pass
         else:
             return False

coremltools/converters/mil/mil/passes/defs/optimize_linear.py

@@ -239,7 +239,7 @@ def _try_to_transform(self, matmul_op, add_op, block):
         d_out = weight.shape[1] if not transpose_weight else weight.shape[0]
         bias = add_op.x.val if add_op.x.val is not None else add_op.y.val
         if len(bias.shape) > 1:
-            if any(d != 1 for d in bias.shape[:-1]):
+            if any([d != 1 for d in bias.shape[:-1]]):
                 return  # cannot transform
 
             # squeeze leading dims of size 1