Replacing assert statements with Exceptions

src/omlt/gbt/gbt_formulation.py

@@ -197,13 +197,23 @@
         node_mask = (nodes_tree_ids == tree_id) & (nodes_node_ids == branch_node_id)
         feature_id = nodes_feature_ids[node_mask]
         branch_value = nodes_values[node_mask]
-        assert len(feature_id) == 1 and len(branch_value) == 1
+        if len(branch_value) != 1:
+            raise ValueError(
+                f"The given tree_id and branch_node_id do not uniquely identify a branch value."
+            )
+        if len(feature_id) != 1:
+            raise ValueError(
+                f"The given tree_id and branch_node_id do not uniquely identify a feature."
+            )
         feature_id = feature_id[0]
         branch_value = branch_value[0]
         (branch_y_idx,) = np.where(
             branch_value_by_feature_id[feature_id] == branch_value
         )
-        assert len(branch_y_idx) == 1
+        if len(branch_y_idx) != 1:
+            raise ValueError(
+                f"The given tree_id and branch_node_id do not uniquely identify a branch index."
+            )
         return block.y[feature_id, branch_y_idx[0]]
 
     def _sum_of_z_l(tree_id, start_node_id):

src/omlt/gbt/model.py

@@ -56,24 +56,36 @@
 def _model_num_inputs(model):
     """Returns the number of input variables"""
     graph = model.graph
-    assert len(graph.input) == 1
+    if len(graph.input) != 1:
+        raise ValueError(
+            f"Model graph input field is multi-valued {graph.input}. A single value is required."
+        )
     return _tensor_size(graph.input[0])
 
 
 def _model_num_outputs(model):
     """Returns the number of output variables"""
     graph = model.graph
-    assert len(graph.output) == 1
+    if len(graph.output) != 1:
+        raise ValueError(
+            f"Model graph output field is multi-valued {graph.output}. A single value is required."
+        )
     return _tensor_size(graph.output[0])
 
 
 def _tensor_size(tensor):
     """Returns the size of an input tensor"""
     tensor_type = tensor.type.tensor_type
     size = None
-    for dim in tensor_type.shape.dim:
-        if dim.dim_value is not None and dim.dim_value > 0:
-            assert size is None
-            size = dim.dim_value
-    assert size is not None
+    dim_values = [
+        dim.dim_value
+        for dim in tensor_type.shape.dim
+        if dim.dim_value is not None and dim.dim_value > 0
+    ]
+    if len(dim_values) == 1:
+        size = dim_values[0]
+    elif dim_values == []:
+        raise ValueError(f"Tensor {tensor} has no positive dimensions.")
+    else:
+        raise ValueError(f"Tensor {tensor} has multiple positive dimensions.")
     return size

src/omlt/io/onnx_parser.py

@@ -73,13 +73,17 @@
                         dim_value = 1
                     size.append(dim.dim_value)
                     dim_value *= dim.dim_value
-            assert dim_value is not None
+            if dim_value is None:
+                raise ValueError(
+                    f'All dimensions in graph "{graph.name}" input tensor have 0 value.'
+                )
             assert network_input is None
             network_input = InputLayer(size)
             self._node_map[input.name] = network_input
             network.add_layer(network_input)
 
-        assert network_input is not None
+        if network_input is None:
+            raise ValueError(f'No valid input layer found in graph "{graph.name}".')
 
         self._nodes = nodes
         self._nodes_by_output = nodes_by_output
@@ -109,11 +113,14 @@
                 # Now connect inputs to the current node
                 for input in node_inputs:
                     self._nodes[input][2].append(node.name)
-            else:
-                assert node.op_type == "Constant"
+            elif node.op_type == "Constant":
                 for output in node.output:
                     value = _parse_constant_value(node)
                     self._constants[output] = value
+            else:
+                raise ValueError(
+                    f'Nodes must have inputs or have op_type "Constant". Node "{node.name}" has no inputs and op_type "{node.op_type}".'
+                )
 
         # traverse graph
         self._node_stack = list(inputs)
@@ -169,34 +176,54 @@
 
     def _consume_dense_nodes(self, node, next_nodes):
         """Starting from a MatMul node, consume nodes to form a dense Ax + b node."""
-        assert node.op_type == "MatMul"
-        assert len(node.input) == 2
+        if node.op_type != "MatMul":
+            raise ValueError(
+                f"{node.name} is a {node.op_type} node, only MatMul nodes can be used as starting points for consumption."
+            )
+        if len(node.input) != 2:
+            raise ValueError(
+                f"{node.name} input has {len(node.input)} dimensions, only nodes with 2 input dimensions can be used as starting points for consumption."
+            )
 
         [in_0, in_1] = list(node.input)
         input_layer, transformer = self._node_input_and_transformer(in_0)
         node_weights = self._initializers[in_1]
 
-        assert len(next_nodes) == 1
+        if len(next_nodes) != 1:
+            raise ValueError(
+                f"Next nodes must have length 1, {next_nodes} has length {len(next_nodes)}"
+            )
 
         # expect 'Add' node ahead
         type_, node, maybe_next_nodes = self._nodes[next_nodes[0]]
-        assert type_ == "node"
-        assert node.op_type == "Add"
+        if type_ != "node":
+            raise TypeError(f"Expected a node next, got a {type_} instead.")
+        if node.op_type != "Add":
+            raise ValueError(
+                f"The first node to be consumed, {node.name}, is a {node.op_type} node. Only Add nodes are supported."
+            )
 
         # extract biases
         next_nodes = maybe_next_nodes
-        assert len(node.input) == 2
         [in_0, in_1] = list(node.input)
 
         if in_0 in self._initializers:
             node_biases = self._initializers[in_0]
-        else:
-            assert in_1 in self._initializers
+        elif in_1 in self._initializers:
             node_biases = self._initializers[in_1]
+        else:
+            raise ValueError(f"Node inputs were not found in graph initializers.")
 
-        assert len(node_weights.shape) == 2
-        assert node_weights.shape[1] == node_biases.shape[0]
-        assert len(node.output) == 1
+        if len(node_weights.shape) != 2:
+            raise ValueError(f"Node weights must be a 2-dimensional matrix.")
+        if node_weights.shape[1] != node_biases.shape[0]:
+            raise ValueError(
+                f"Node weights has {node_weights.shape[1]} columns; node biases has {node_biases.shape[0]} rows. These must be equal."
+            )
+        if len(node.output) != 1:
+            raise ValueError(
+                f"Node output is {node.output} but should be a single value."
+            )
 
         input_output_size = _get_input_output_size(input_layer, transformer)
 
@@ -226,8 +253,14 @@
 
     def _consume_gemm_dense_nodes(self, node, next_nodes):
         """Starting from a Gemm node, consume nodes to form a dense aAB + bC node."""
-        assert node.op_type == "Gemm"
-        assert len(node.input) == 3
+        if node.op_type != "Gemm":
+            raise ValueError(
+                f"{node.name} is a {node.op_type} node, only Gemm nodes can be used as starting points for consumption."
+            )
+        if len(node.input) != 3:
+            raise ValueError(
+                f"{node.name} input has {len(node.input)} dimensions, only nodes with 3 input dimensions can be used as starting points for consumption."
+            )
 
         attr = _collect_attributes(node)
         alpha = attr["alpha"]
@@ -275,8 +308,15 @@
         Starting from a Conv node, consume nodes to form a convolution node with
         (optional) activation function.
         """
-        assert node.op_type == "Conv"
-        assert len(node.input) in [2, 3]
+        if node.op_type != "Conv":
+            raise ValueError(
+                f"{node.name} is a {node.op_type} node, only Conv nodes can be used as starting points for consumption."
+            )
+        if len(node.input) not in [2, 3]:
+            raise ValueError(
+                f"{node.name} input has {len(node.input)} dimensions, only nodes with 2 or 3 input dimensions can be used as starting points for consumption."
+            )
+
         if len(node.input) == 2:
             [in_0, in_1] = list(node.input)
             in_2 = None
@@ -295,18 +335,43 @@
         attr = _collect_attributes(node)
 
         strides = attr["strides"]
-
         # check only kernel shape and stride are set
-        # everything else is not supported
-        assert biases.shape == (out_channels,)
-        assert in_channels == input_output_size[0]
-        assert attr["kernel_shape"] == kernel_shape
-        assert attr["dilations"] == [1, 1]
-        assert attr["group"] == 1
-        if "pads" in attr:
-            assert not np.any(attr["pads"])  # pads all zero
-        assert len(kernel_shape) == len(strides)
-        assert len(input_output_size) == len(kernel_shape) + 1
+        if attr["kernel_shape"] != kernel_shape:
+            raise ValueError(
+                f"Kernel shape attribute {attr['kernel_shape']} does not match initialized kernel shape {kernel_shape}."
+            )
+        if len(kernel_shape) != len(strides):
+            raise ValueError(
+                f"Initialized kernel shape {kernel_shape} has {len(kernel_shape)} dimensions. Strides attribute has {len(strides)} dimensions. These must be equal."
+            )
+        if len(input_output_size) != len(kernel_shape) + 1:
+            raise ValueError(
+                f"Input/output size ({input_output_size}) must have one more dimension than initialized kernel shape ({kernel_shape})."
+            )
+
+        # Check input, output have correct dimensions
+        if biases.shape != (out_channels,):
+            raise ValueError(
+                f"Biases shape {biases.shape} must match output weights channels {(out_channels,)}."
+            )
+        if in_channels != input_output_size[0]:
+            raise ValueError(
+                f"Input/output size ({input_output_size}) first dimension must match input weights channels ({in_channels})."
+            )
+
+        # Other attributes are not supported
+        if "dilations" in attr and attr["dilations"] != [1, 1]:
+            raise ValueError(
+                f"{node} has non-identity dilations ({attr['dilations']}). This is not supported."
+            )
+        if attr["group"] != 1:
+            raise ValueError(
+                f"{node} has multiple groups ({attr['group']}). This is not supported."
+            )
+        if "pads" in attr and np.any(attr["pads"]):
+            raise ValueError(
+                f"{node} has non-zero pads ({attr['pads']}). This is not supported."
+            )
 
         # generate new nodes for the node output
         padding = 0
@@ -326,7 +391,10 @@
 
         # convolute image one channel at the time
         # expect 2d image with channels
-        assert len(input_output_size) == 3
+        if len(input_output_size) != 3:
+            raise ValueError(
+                f"Expected a 2D image with channels, got {input_output_size}."
+            )
 
         conv_layer = ConvLayer2D(
             input_output_size,
@@ -343,8 +411,14 @@
 
     def _consume_reshape_nodes(self, node, next_nodes):
         """Parse a Reshape node."""
-        assert node.op_type == "Reshape"
-        assert len(node.input) == 2
+        if node.op_type != "Reshape":
+            raise ValueError(
+                f"{node.name} is a {node.op_type} node, only Reshape nodes can be used as starting points for consumption."
+            )
+        if len(node.input) != 2:
+            raise ValueError(
+                f"{node.name} input has {len(node.input)} dimensions, only nodes with 2 input dimensions can be used as starting points for consumption."
+            )
         [in_0, in_1] = list(node.input)
         input_layer = self._node_map[in_0]
         new_shape = self._constants[in_1]
@@ -358,23 +432,34 @@
         Starting from a MaxPool node, consume nodes to form a pooling node with
         (optional) activation function.
         """
-        assert node.op_type in _POOLING_OP_TYPES
+        if node.op_type not in _POOLING_OP_TYPES:
+            raise ValueError(
+                f"{node.name} is a {node.op_type} node, only MaxPool nodes can be used as starting points for consumption."
+            )
         pool_func_name = "max"
 
         # ONNX network should not contain indices output from MaxPool - not supported by OMLT
-        assert len(node.output) == 1
+        if len(node.output) != 1:
+            raise ValueError(
+                f"The ONNX contains indices output from MaxPool. This is not supported by OMLT."
+            )
+        if len(node.input) != 1:
+            raise ValueError(
+                f"{node.name} input has {len(node.input)} dimensions, only nodes with 1 input dimension can be used as starting points for consumption."
+            )
 
-        assert len(node.input) == 1
         input_layer, transformer = self._node_input_and_transformer(node.input[0])
         input_output_size = _get_input_output_size(input_layer, transformer)
 
         # currently only support 2D image with channels.
         if len(input_output_size) == 4:
             # this means there is an extra dimension for number of batches
             # batches not supported, so only accept if they're not there or there is only 1 batch
-            assert input_output_size[0] == 1
+            if input_output_size[0] != 1:
+                raise ValueError(
+                    f"{node.name} has {input_output_size[0]} batches, only a single batch is supported."
+                )
             input_output_size = input_output_size[1:]
-        assert len(input_output_size) == 3
 
         in_channels = input_output_size[0]
 
@@ -385,11 +470,26 @@
 
         # check only kernel shape, stride, storage order are set
         # everything else is not supported
-        assert ("dilations" not in attr) or (attr["dilations"] == [1, 1])
-        assert ("pads" not in attr) or (not np.any(attr["pads"]))
-        assert ("auto_pad" not in attr) or (attr["auto_pad"] == "NOTSET")
-        assert len(kernel_shape) == len(strides)
-        assert len(input_output_size) == len(kernel_shape) + 1
+        if "dilations" in attr and attr["dilations"] != [1, 1]:
+            raise ValueError(
+                f"{node.name} has non-identity dilations ({attr['dilations']}). This is not supported."
+            )
+        if "pads" in attr and np.any(attr["pads"]):
+            raise ValueError(
+                f"{node.name} has non-zero pads ({attr['pads']}). This is not supported."
+            )
+        if ("auto_pad" in attr) and (attr["auto_pad"] != "NOTSET"):
+            raise ValueError(
+                f"{node.name} has autopad set ({attr['auto_pad']}). This is not supported."
+            )
+        if len(kernel_shape) != len(strides):
+            raise ValueError(
+                f"Kernel shape {kernel_shape} has {len(kernel_shape)} dimensions. Strides attribute has {len(strides)} dimensions. These must be equal."
+            )
+        if len(input_output_size) != len(kernel_shape) + 1:
+            raise ValueError(
+                f"Input/output size ({input_output_size}) must have one more dimension than kernel shape ({kernel_shape})."
+            )
 
         output_shape_wrapper = math.floor
         if "ceil_mode" in attr and attr["ceil_mode"] == 1: