Merge pull request #216 from igabirondo16/feature/pytorch_geometric

Feature/pytorch geometric

Author: ChrisCummins

programl/__init__.py

@@ -52,7 +52,7 @@
     to_bytes,
     to_string,
 )
-from programl.transform_ops import to_dgl, to_dot, to_json, to_networkx
+from programl.transform_ops import to_dgl, to_dot, to_json, to_networkx, to_pyg
 from programl.util.py.runfiles_path import runfiles_path
 from programl.version import PROGRAML_VERSION
 
@@ -84,6 +84,7 @@
     "to_dot",
     "to_json",
     "to_networkx",
+    "to_pyg",
     "to_string",
     "UnsupportedCompiler",
 ]

programl/requirements.txt

@@ -5,4 +5,5 @@ networkx>=2.4
 numpy>=1.19.3
 protobuf>=3.13.0,<4.21.0
 torch>=1.8.0
+torch_geometric==2.4.0
 tqdm>=4.38.0

programl/transform_ops.py

@@ -18,11 +18,13 @@
 """
 import json
 import subprocess
+import torch
 from typing import Any, Dict, Iterable, Optional, Union
 
 import dgl
 import networkx as nx
 from dgl.heterograph import DGLHeteroGraph
+from torch_geometric.data import HeteroData
 from networkx.readwrite import json_graph as nx_json
 
 from programl.exceptions import GraphTransformError
@@ -258,3 +260,122 @@ def _run_one(graph: ProgramGraph) -> str:
     if isinstance(graphs, ProgramGraph):
         return _run_one(graphs)
     return execute(_run_one, graphs, executor, chunksize)
+
+
+def to_pyg(
+    graphs: Union[ProgramGraph, Iterable[ProgramGraph]],
+    timeout: int = 300,
+    vocabulary: Optional[Dict[str, int]] = None,
+    executor: Optional[ExecutorLike] = None,
+    chunksize: Optional[int] = None,
+) -> Union[HeteroData, Iterable[HeteroData]]:
+    """Convert one or more Program Graphs to Pytorch-Geometrics's HeteroData.
+    This graphs can be used as input for any deep learning model built with
+    Pytorch-Geometric:
+
+    https://pytorch-geometric.readthedocs.io/en/latest/tutorial/heterogeneous.html
+
+    :param graphs: A Program Graph, or a sequence of Program Graphs.
+
+    :param timeout: The maximum number of seconds to wait for an individual
+        graph conversion before raising an error. If multiple inputs are
+        provided, this timeout is per-input.
+
+    :param vocabulary: A dictionary containing ProGraML's vocabulary, where the
+        keys are the text attribute of the nodes and the values their respective
+        indexes.
+
+    :param executor: An executor object, with method :code:`submit(callable,
+        *args, **kwargs)` and returning a Future-like object with methods
+        :code:`done() -> bool` and :code:`result() -> float`. The executor role
+        is to dispatch the execution of the jobs locally/on a cluster/with
+        multithreading depending on the implementation. Eg:
+        :code:`concurrent.futures.ThreadPoolExecutor`. Defaults to single
+        threaded execution. This is only used when multiple inputs are given.
+
+    :param chunksize: The number of inputs to read and process at a time. A
+        larger chunksize improves parallelism but increases memory consumption
+        as more inputs must be stored in memory. This is only used when multiple
+        inputs are given.
+
+    :return: A HeteroData graph when a single input is provided, else an
+        iterable sequence of HeteroData graphs.
+    """
+
+    def _run_one(graph: ProgramGraph) -> HeteroData:
+        # 4 lists, one per edge type
+        # (control, data, call and type edges)
+        adjacencies = [[], [], [], []]
+        edge_positions = [[], [], [], []]
+
+        # Create the adjacency lists and the positions
+        for edge in graph.edge:
+            adjacencies[edge.flow].append([edge.source, edge.target])
+            edge_positions[edge.flow].append(edge.position)
+
+        # Store the text attributes
+        node_text_list = []
+        node_full_text_list = []
+
+        # Store the text and full text attributes
+        for node in graph.node:
+            node_text = node_full_text = node.text
+
+            if (
+                node.features
+                and node.features.feature["full_text"].bytes_list.value
+            ):
+                node_full_text = node.features.feature["full_text"].bytes_list.value[0]
+
+            node_text_list.append(node_text)
+            node_full_text_list.append(node_full_text)
+
+
+        vocab_ids = None
+        if vocabulary is not None:
+            vocab_ids = [
+                vocabulary.get(node.text, len(vocabulary.keys()))
+                for node in graph.node
+            ]
+
+        # Pass from list to tensor
+        adjacencies = [torch.tensor(adj_flow_type) for adj_flow_type in adjacencies]
+        edge_positions = [torch.tensor(edge_pos_flow_type) for edge_pos_flow_type in edge_positions]
+
+        if vocabulary is not None:
+            vocab_ids = torch.tensor(vocab_ids)
+
+        # Create the graph structure
+        hetero_graph = HeteroData()
+
+        # Vocabulary index of each node
+        hetero_graph['nodes']['text'] = node_text_list
+        hetero_graph['nodes']['full_text'] = node_full_text_list
+        hetero_graph['nodes'].x = vocab_ids
+
+        # Add the adjacency lists
+        hetero_graph['nodes', 'control', 'nodes'].edge_index = (
+            adjacencies[0].t().contiguous() if adjacencies[0].nelement() > 0 else torch.tensor([[], []])
+        )
+        hetero_graph['nodes', 'data', 'nodes'].edge_index = (
+            adjacencies[1].t().contiguous() if adjacencies[1].nelement() > 0 else torch.tensor([[], []])
+        )
+        hetero_graph['nodes', 'call', 'nodes'].edge_index = (
+            adjacencies[2].t().contiguous() if adjacencies[2].nelement() > 0 else torch.tensor([[], []])
+        )
+        hetero_graph['nodes', 'type', 'nodes'].edge_index = (
+            adjacencies[3].t().contiguous() if adjacencies[3].nelement() > 0 else torch.tensor([[], []])
+        )
+
+        # Add the edge positions
+        hetero_graph['nodes', 'control', 'nodes'].edge_attr = edge_positions[0]
+        hetero_graph['nodes', 'data', 'nodes'].edge_attr = edge_positions[1]
+        hetero_graph['nodes', 'call', 'nodes'].edge_attr = edge_positions[2]
+        hetero_graph['nodes', 'type', 'nodes'].edge_attr = edge_positions[3]
+
+        return hetero_graph
+
+    if isinstance(graphs, ProgramGraph):
+        return _run_one(graphs)
+
+    return execute(_run_one, graphs, executor, chunksize)

tests/BUILD

@@ -134,3 +134,14 @@ py_test(
         "//tests/plugins",
     ],
 )
+
+py_test(
+    name = "to_pyg_test",
+    srcs = ["to_pyg_test.py"],
+    shard_count = 8,
+    deps = [
+        "//programl",
+        "//tests:test_main",
+        "//tests/plugins",
+    ],
+)

tests/to_pyg_test.py

@@ -0,0 +1,122 @@
+# Copyright 2019-2020 the ProGraML authors.
+#
+# Contact Chris Cummins <chrisc.101@gmail.com>.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from concurrent.futures.thread import ThreadPoolExecutor
+
+import networkx as nx
+import pytest
+
+import programl as pg
+from torch_geometric.data import HeteroData
+from tests.test_main import main
+
+pytest_plugins = ["tests.plugins.llvm_program_graph"]
+
+
+@pytest.fixture(scope="session")
+def graph() -> pg.ProgramGraph:
+    return pg.from_cpp("int A() { return 0; }")
+
+@pytest.fixture(scope="session")
+def graph2() -> pg.ProgramGraph:
+    return pg.from_cpp("int B() { return 1; }")
+
+@pytest.fixture(scope="session")
+def graph3() -> pg.ProgramGraph:
+    return pg.from_cpp("int B(int x) { return x + 1; }")
+
+def graphs_are_equal(
+    graph1: HeteroData,
+    graph2: HeteroData,
+):
+    return (
+        (graph1['nodes']['full_text'] == graph2['nodes']['full_text'])
+        and (graph1['nodes', 'control', 'nodes'].edge_index.equal(graph2['nodes', 'control', 'nodes'].edge_index))
+        and (graph1['nodes', 'data', 'nodes'].edge_index.equal(graph2['nodes', 'data', 'nodes'].edge_index))
+        and (graph1['nodes', 'call', 'nodes'].edge_index.equal(graph2['nodes', 'call', 'nodes'].edge_index))
+        and (graph1['nodes', 'type', 'nodes'].edge_index.equal(graph2['nodes', 'type', 'nodes'].edge_index))
+    )
+
+def test_to_pyg_simple_graph(graph: pg.ProgramGraph):
+    graphs = list(pg.to_pyg([graph]))
+    assert len(graphs) == 1
+    assert isinstance(graphs[0], HeteroData)
+
+def test_to_pyg_simple_graph_single_input(graph: pg.ProgramGraph):
+    pyg_graph = pg.to_pyg(graph)
+    assert isinstance(pyg_graph, HeteroData)
+
+def test_to_pyg_different_two_different_inputs(
+    graph: pg.ProgramGraph,
+    graph2: pg.ProgramGraph,
+):
+    pyg_graph = pg.to_pyg(graph)
+    pyg_graph2 = pg.to_pyg(graph2)
+
+    # Ensure that the graphs are different
+    assert not graphs_are_equal(pyg_graph, pyg_graph2)
+
+def test_to_pyg_different_inputs(
+    graph: pg.ProgramGraph,
+    graph2: pg.ProgramGraph,
+    graph3: pg.ProgramGraph
+):
+    pyg_graph = pg.to_pyg(graph)
+    pyg_graph2 = pg.to_pyg(graph2)
+    pyg_graph3 = pg.to_pyg(graph3)
+
+    # Ensure that the graphs are different
+    assert not graphs_are_equal(pyg_graph, pyg_graph2)
+    assert not graphs_are_equal(pyg_graph, pyg_graph3)
+    assert not graphs_are_equal(pyg_graph2, pyg_graph3)
+
+def test_to_pyg_two_inputs(graph: pg.ProgramGraph):
+    graphs = list(pg.to_pyg([graph, graph]))
+    assert len(graphs) == 2
+    assert graphs_are_equal(graphs[0], graphs[1])
+
+def test_to_pyg_generator(graph: pg.ProgramGraph):
+    graphs = list(pg.to_pyg((graph for _ in range(10)), chunksize=3))
+    assert len(graphs) == 10
+    for x in graphs[1:]:
+        assert  graphs_are_equal(graphs[0], x)
+
+def test_to_pyg_generator_parallel_executor(graph: pg.ProgramGraph):
+    with ThreadPoolExecutor() as executor:
+        graphs = list(
+            pg.to_pyg((graph for _ in range(10)), chunksize=3, executor=executor)
+        )
+    assert len(graphs) == 10
+    for x in graphs[1:]:
+        assert graphs_are_equal(graphs[0], x)
+
+def test_to_pyg_smoke_test(llvm_program_graph: pg.ProgramGraph):
+    graphs = list(pg.to_pyg([llvm_program_graph]))
+
+    num_nodes = len(graphs[0]['nodes']['text'])
+    num_control_edges = graphs[0]['nodes', 'control', 'nodes'].edge_index.size(1)
+    num_data_edges = graphs[0]['nodes', 'data', 'nodes'].edge_index.size(1)
+    num_call_edges = graphs[0]['nodes', 'call', 'nodes'].edge_index.size(1)
+    num_type_edges = graphs[0]['nodes', 'type', 'nodes'].edge_index.size(1)
+    num_edges = num_control_edges + num_data_edges + num_call_edges + num_type_edges
+
+    assert len(graphs) == 1
+    assert isinstance(graphs[0], HeteroData)
+    assert num_nodes == len(llvm_program_graph.node)
+    assert num_edges <= len(llvm_program_graph.edge)
+
+
+if __name__ == "__main__":
+    main()