Support kotgll evaluation

Author: IlyaMuravjov

cfpq_eval/runners/all_pairs_cflr_tool_runner_facade.py

@@ -6,6 +6,8 @@
     GigascaleAllPairsCflrToolRunner)
 from cfpq_eval.runners.graspan_algo_all_pairs_cflr_tool_runner import (
     GraspanAllPairsCflrToolRunner)
+from cfpq_eval.runners.kotgll_all_pairs_cflr_tool_runner import (
+    KotgllAllPairsCflrToolRunner)
 from cfpq_eval.runners.pearl_algo_all_pairs_cflr_tool_runner import (
     PearlAllPairsCflrToolRunner)
 from cfpq_eval.runners.pocr_algo_all_pairs_cflr_tool_runner import (
@@ -24,7 +26,8 @@ def run_appropriate_all_pairs_cflr_tool(
         "pocr": PocrAllPairsCflrToolRunner,
         "pearl": PearlAllPairsCflrToolRunner,
         "gigascale": GigascaleAllPairsCflrToolRunner,
-        "graspan": GraspanAllPairsCflrToolRunner
+        "graspan": GraspanAllPairsCflrToolRunner,
+        "kotgll": KotgllAllPairsCflrToolRunner,
     }.get(algo_settings, PyAlgoAllPairsCflrToolRunner)(
         algo_settings, graph_path, grammar_path, timeout_sec
     ).run()

cfpq_eval/runners/kotgll_all_pairs_cflr_tool_runner.py

@@ -0,0 +1,105 @@
+import os
+import subprocess
+import uuid
+import warnings
+from pathlib import Path
+from typing import Optional
+
+import pandas as pd
+
+from cfpq_eval.runners.all_pairs_cflr_tool_runner import (
+    AbstractAllPairsCflrToolRunner,
+    CflrToolRunResult
+)
+from cfpq_model.cnf_grammar_template import CnfGrammarTemplate
+from cfpq_model.label_decomposed_graph import LabelDecomposedGraph
+from cfpq_model.model_utils import explode_indices
+
+
+class KotgllAllPairsCflrToolRunner(AbstractAllPairsCflrToolRunner):
+    @property
+    def base_command(self) -> Optional[str]:
+        grammar = CnfGrammarTemplate.read_from_pocr_cnf_file(self.grammar_path)
+        graph = LabelDecomposedGraph.read_from_pocr_graph_file(self.graph_path)
+
+        grammar_path = self.grammar_path.parent / "kotgll" / (self.grammar_path.stem + ".rsm")
+        if not os.path.isfile(grammar_path):
+            grammar_path = grammar_path.with_suffix(".cfg")
+            if not os.path.isfile(grammar_path):
+                warnings.warn(
+                    "Skipping kotgll evaluation, because RSM/CFG is missing. "
+                    "To fix this error write RSM/CFG to "
+                    f"'{grammar_path.with_suffix('.rsm')}'/'{grammar_path.with_suffix('.cfg')}'. "
+                    "See https://github.com/vadyushkins/kotgll?tab=readme-ov-file#rsm-format-example"
+                )
+                return None
+        if graph.block_matrix_space.block_count > 1:
+            exploded_grammar_path = (
+                    self.grammar_path.parent / "kotgll" / self.grammar_path.stem /
+                    self.graph_path.stem / (self.grammar_path.stem + grammar_path.suffix)
+            )
+            self._explode_grammar(
+                grammar_path,
+                exploded_grammar_path,
+                graph.block_matrix_space.block_count
+            )
+            grammar_path = exploded_grammar_path
+
+        # kotgll doesn't support indexed symbols, we need to concat labels and indices
+        graph, grammar = explode_indices(graph, grammar)
+        graph_path = self.graph_path.parent / "kotgll" / self.graph_path.stem / self.graph_path.name
+        os.makedirs(graph_path.parent, exist_ok=True)
+        # kotgll requires its own graph formatting, so we need to save the graph in a custom format
+        self._write_kotgll_graph(graph, graph_path)
+
+        out_folder = self.graph_path.parent / "kotgll" / 'out' / str(uuid.uuid4())
+        os.makedirs(out_folder)
+
+        return (
+            'java -cp kotgll.jar org.kotgll.benchmarks.BenchmarksKt '
+            f'--grammar {grammar_path.suffix[1:]} --sppf off '
+            f'--inputPath {graph_path.parent} --grammarPath {grammar_path} '
+            f'--outputPath {out_folder} '
+            '--warmUpRounds 1 --benchmarkRounds 1'
+        )
+
+    @property
+    def work_dir(self) -> Optional[Path]:
+        return Path(os.environ['KOTGLL_DIR'])
+
+    def parse_results(self, process: subprocess.CompletedProcess[str]) -> CflrToolRunResult:
+        for line in process.stdout.split('\n'):
+            if line.startswith('benchmark::'):
+                parts = line.split()
+                return CflrToolRunResult(
+                    s_edges=int(parts[-2]),
+                    time_sec=float(parts[-1]),
+                    ram_kb=self.parse_ram_usage_kb(process)
+                )
+        raise Exception(f"No results are found in stdout {process.stdout}")
+
+    @staticmethod
+    def _write_kotgll_graph(graph: LabelDecomposedGraph, graph_path: Path) -> None:
+        with open(graph_path, 'w', encoding="utf-8") as output_file:
+            for symbol, matrix in graph.matrices.items():
+                edge_label = symbol.label
+                (rows, columns, _) = matrix.to_coo()
+                edges_df = pd.DataFrame({
+                    'source': rows,
+                    'destination': columns,
+                    'label': edge_label
+                })
+                csv_string = edges_df.to_csv(sep=' ', index=False, header=False)
+                output_file.write(csv_string)
+
+    @staticmethod
+    def _explode_grammar(rsm_path: Path, exploded_grammar_path: Path, block_count: int) -> None:
+        os.makedirs(exploded_grammar_path.parent, exist_ok=True)
+        with open(rsm_path, 'r') as infile, open(exploded_grammar_path, 'w') as outfile:
+            for line in infile:
+                if '{' in line and '}' in line:
+                    for i in range(block_count + 1):
+                        expanded_line = eval(f"f'{line.strip()}'")
+                        outfile.write(expanded_line + '\n')
+                else:
+                    outfile.write(line)