Source code for onnx_array_api.plotting.dot_plot

import re
from typing import Dict, Optional, Tuple

from onnx import GraphProto, ModelProto
from onnx.helper import tensor_dtype_to_string

from ..reference import to_array_extended as to_array
from ._helper import Graph, _get_shape, attributes_as_dict


def _type_to_string(dtype):
    """
    Converts a type into a readable string.
    """
    if dtype.HasField("tensor_type"):
        ttype = dtype.tensor_type
        return tensor_dtype_to_string(ttype.elem_type)
    if dtype.HasField("sequence_type"):
        stype = dtype.sequence_type
        return f"Sequence[{type(stype.elem_type)}]"
    raise ValueError(f"Unable to convert {dtype} into a string.")


[docs]def to_dot( proto: ModelProto, recursive: bool = False, prefix: str = "", use_onnx: bool = False, add_functions: bool = True, rt_shapes: Optional[Dict[str, Tuple[int, ...]]] = None, **params, ) -> str: """ Produces a :epkg:`DOT` language string for the graph. :param params: additional params to draw the graph :param recursive: also show subgraphs inside operator like `Scan` :param prefix: prefix for every node name :param use_onnx: use :epkg:`onnx` dot format instead of this one :param add_functions: add functions to the graph :param rt_shapes: indicates shapes obtained from the execution or inference :return: string Default options for the graph are: :: options = { 'orientation': 'portrait', 'ranksep': '0.25', 'nodesep': '0.05', 'width': '0.5', 'height': '0.1', 'size': '7', } One example: .. runpython:: :showcode: :warningout: DeprecationWarning, FutureWarning :process: import numpy as np # B from onnx_array_api.npx import absolute, jit_onnx from onnx_array_api.plotting.dot_plot import to_dot def l1_loss(x, y): return absolute(x - y).sum() def l2_loss(x, y): return ((x - y) ** 2).sum() def myloss(x, y): return l1_loss(x[:, 0], y[:, 0]) + l2_loss(x[:, 1], y[:, 1]) jitted_myloss = jit_onnx(myloss) x = np.array([[0.1, 0.2], [0.3, 0.4]], dtype=np.float32) y = np.array([[0.11, 0.22], [0.33, 0.44]], dtype=np.float32) res = jitted_myloss(x, y) print(res) .. gdot:: :script: DOT-SECTION :process: # to_dot import numpy as np from onnx_array_api.npx import absolute, jit_onnx from onnx_array_api.plotting.dot_plot import to_dot def l1_loss(x, y): return absolute(x - y).sum() def l2_loss(x, y): return ((x - y) ** 2).sum() def myloss(x, y): return l1_loss(x[:, 0], y[:, 0]) + l2_loss(x[:, 1], y[:, 1]) jitted_myloss = jit_onnx(myloss) x = np.array([[0.1, 0.2], [0.3, 0.4]], dtype=np.float32) y = np.array([[0.11, 0.22], [0.33, 0.44]], dtype=np.float32) res = jitted_myloss(x, y) print(to_dot(jitted_myloss.get_onnx())) """ clean_label_reg1 = re.compile("\\\\x\\{[0-9A-F]{1,6}\\}") clean_label_reg2 = re.compile("\\\\p\\{[0-9P]{1,6}\\}") def dot_name(text): return text.replace("/", "_").replace(":", "__").replace(".", "_") def dot_label(text): if text is None: return "" for reg in [clean_label_reg1, clean_label_reg2]: fall = reg.findall(text) for f in fall: text = text.replace(f, "_") return text options = { "orientation": "portrait", "ranksep": "0.25", "nodesep": "0.05", "width": "0.5", "height": "0.1", "size": "7", } options.update({k: v for k, v in params.items() if v is not None}) if use_onnx: from onnx.tools.net_drawer import GetOpNodeProducer, GetPydotGraph pydot_graph = GetPydotGraph( proto.graph, name=proto.graph.name, rankdir=params.get("rankdir", "TB"), node_producer=GetOpNodeProducer( "docstring", fillcolor="orange", style="filled", shape="box" ), ) return pydot_graph.to_string() inter_vars = {} exp = ["digraph{"] for opt in {"orientation", "pad", "nodesep", "ranksep", "size"}: if opt in options: exp.append(f" {opt}={options[opt]};") fontsize = 10 shapes = {} if rt_shapes: for name, shape in rt_shapes.items(): va = str(shape.shape) shapes[name] = va # inputs exp.append("") graph = proto.graph if isinstance(proto, ModelProto) else proto for obj in graph.input: if isinstance(obj, str): exp.append( ' {2}{0} [shape=box color=red label="{0}" fontsize={1}];' "".format(obj, fontsize, prefix) ) inter_vars[obj] = obj else: sh = _get_shape(obj) if sh: sh = f"\\nshape={sh}" exp.append( ' {3}{0} [shape=box color=red label="{0}\\n{1}{4}" fontsize={2}];' "".format( obj.name, _type_to_string(obj.type), fontsize, prefix, dot_label(sh) ) ) inter_vars[obj.name] = obj # outputs exp.append("") for obj in graph.output: if isinstance(obj, str): exp.append( ' {2}{0} [shape=box color=green label="{0}" fontsize={1}];'.format( obj, fontsize, prefix ) ) inter_vars[obj] = obj else: sh = _get_shape(obj) if sh: sh = f"\\nshape={sh}" exp.append( f" {prefix}{obj.name} [shape=box color=green " f'label="{obj.name}\\n{_type_to_string(obj.type)}' f'{dot_label(sh)}" fontsize={fontsize}];' ) inter_vars[obj.name] = obj # initializer exp.append("") if hasattr(proto, "graph"): inits = list(proto.graph.initializer) + list(proto.graph.sparse_initializer) for obj in inits: val = to_array(obj) flat = val.flatten() if flat.shape[0] < 9: st = str(val) else: st = str(val) if len(st) > 50: st = st[:50] + "..." st = st.replace("\n", "\\n") kind = "" exp.append( f" {prefix}{dot_name(obj.name)} " f'[shape=box label="{dot_name(obj.name)}' f"\\n{kind}{val.dtype}({val.shape})" f'\\n{dot_label(st)}" fontsize={fontsize}];' ) inter_vars[obj.name] = obj # nodes fill_names = {} if hasattr(proto, "graph"): static_inputs = [n.name for n in proto.graph.input] static_inputs.extend(n.name for n in proto.graph.initializer) static_inputs.extend(n.name for n in proto.graph.sparse_initializer) nodes = list(proto.graph.node) else: static_inputs = list(proto.input) nodes = proto.node for node in nodes: exp.append("") for out in node.output: if out and out not in inter_vars: inter_vars[out] = out sh = shapes.get(out, "") if sh: sh = f"\\nshape={sh}" exp.append( ' {2}{0} [shape=box label="{0}{3}" fontsize={1}];'.format( dot_name(out), fontsize, dot_name(prefix), dot_label(sh) ) ) static_inputs.append(out) if node.name.strip() == "" or node.name in fill_names: name = node.op_type iname = 1 while name in fill_names: name = "%s%d" % (name, iname) iname += 1 node.name = name fill_names[name] = node atts = [] node_attributes = attributes_as_dict(node) for k, v in sorted(node_attributes.items()): if isinstance(v, (GraphProto, Graph)): continue val = str(v).replace("\n", "\\n").replace('"', "'") sl = max(30 - len(k), 10) if len(val) > sl: val = val[:sl] + "..." if val is not None: atts.append(f"{k}={val}") satts = "" if len(atts) == 0 else ("\\n" + "\\n".join(atts)) connects = [] if recursive and node.op_type in {"Scan", "Loop", "If"}: fields = ( ["then_branch", "else_branch"] if node.op_type == "If" else ["body"] ) for field in fields: if field not in node_attributes: continue # creates the subgraph body = node_attributes[field] subprefix = prefix + "B_" subdot = to_dot( body, recursive=recursive, prefix=subprefix, rt_shapes=rt_shapes ) lines = subdot.split("\n") start = 0 for i, line in enumerate(lines): if "[" in line: start = i break subgraph = "\n".join(lines[start:]) # connecting the subgraph cluster = f"cluster_{node.op_type}{id(node)}_{id(field)}" exp.append(f" subgraph {cluster} {{") exp.append(f' label="{node.op_type}\\n({dot_name(field)}){satts}";') exp.append(f" fontsize={fontsize};") exp.append(" color=black;") exp.append("\n".join(map(lambda s: " " + s, subgraph.split("\n")))) node0 = body.node[0] connects.append( (f"{dot_name(subprefix)}{dot_name(node0.name)}", cluster) ) for inp1, inp2 in zip(node.input, body.input): exp.append( f" {dot_name(prefix)}{dot_name(inp1)} -> " f"{dot_name(subprefix)}{dot_name(inp2.name)};" ) for out1, out2 in zip(body.output, node.output): if not out2: # Empty output, it cannot be used. continue exp.append( f" {dot_name(subprefix)}{dot_name(out1.name)} -> " f"{dot_name(prefix)}{dot_name(out2)};" ) else: exp.append( f" {dot_name(prefix)}{dot_name(node.name)} " f'[shape=box style="filled,rounded" color=orange ' f'label="{node.op_type}{satts}" ' f"fontsize={fontsize}];" ) if connects is not None and len(connects) > 0: for name, cluster in connects: exp.append( f" {dot_name(prefix)}{dot_name(node.name)} -> " f"{name} [lhead={cluster}];" ) for inp in node.input: exp.append( f" {dot_name(prefix)}{dot_name(inp)} -> " f"{dot_name(prefix)}{dot_name(node.name)};" ) for out in node.output: if not out: # Empty output, it cannot be used. continue exp.append( f" {dot_name(prefix)}{dot_name(node.name)} -> " f"{dot_name(prefix)}{dot_name(out)};" ) functions = getattr(proto, "function", []) if add_functions and len(functions) > 0: for f in functions: dot = to_dot( f, recursive=recursive, prefix=prefix + f.name, use_onnx=use_onnx, add_functions=False, rt_shapes=rt_shapes, **params, ) spl = dot.split("\n")[1:] exp.append("") exp.append(" subgraph cluster_%d {" % i) exp.append(f' label="{v.obj.name}";') exp.append(" color=blue;") # exp.append(' style=filled;') exp.extend((" " + line) for line in spl) exp.append("}") return "\n".join(exp)