Coverage for sparkle/solver/selector.py: 59%
64 statements
« prev ^ index » next coverage.py v7.6.4, created at 2024-11-05 14:48 +0000
1"""File to handle a Selector for selecting Solvers."""
2from __future__ import annotations
3from pathlib import Path
4import subprocess
5import ast
7import runrunner as rrr
8from runrunner import Runner, Run
10from sparkle.types import SparkleCallable, SparkleObjective
11from sparkle.structures import FeatureDataFrame, PerformanceDataFrame
14class Selector(SparkleCallable):
15 """The Selector class for handling Algorithm Selection."""
17 def __init__(self: SparkleCallable,
18 executable_path: Path,
19 raw_output_directory: Path) -> None:
20 """Initialize the Selector object.
22 Args:
23 executable_path: Path of the Selector executable.
24 raw_output_directory: Directory where the Selector will write its raw output.
25 Defaults to directory / tmp
26 """
27 self.selector_builder_path = executable_path
28 self.directory = self.selector_builder_path.parent
29 self.name = self.selector_builder_path.name
30 self.raw_output_directory = raw_output_directory
32 if not self.raw_output_directory.exists():
33 self.raw_output_directory.mkdir(parents=True)
35 def build_construction_cmd(
36 self: Selector,
37 target_file: Path,
38 performance_data: Path,
39 feature_data: Path,
40 objective: SparkleObjective,
41 runtime_cutoff: int | float | str = None,
42 wallclock_limit: int | float | str = None) -> list[str | Path]:
43 """Builds the commandline call string for constructing the Selector.
45 Args:
46 target_file: Path to the file to save the Selector to.
47 performance_data: Path to the performance data csv.
48 feature_data: Path to the feature data csv.
49 objective: The objective to optimize for selection.
50 runtime_cutoff: Cutoff for the runtime in seconds. Defaults to None
51 wallclock_limit: Cutoff for total wallclock in seconds. Defaults to None
53 Returns:
54 The command list for constructing the Selector.
55 """
56 objective_function = "runtime" if objective.time else "solution_quality"
57 # Python3 to avoid execution rights
58 cmd = ["python3", self.selector_builder_path,
59 "--performance_csv", performance_data,
60 "--feature_csv", feature_data,
61 "--objective", objective_function,
62 "--save", target_file]
63 if runtime_cutoff is not None:
64 cmd.extend(["--runtime_cutoff", str(runtime_cutoff), "--tune"])
65 if wallclock_limit is not None:
66 cmd.extend(["--wallclock_limit", str(wallclock_limit)])
67 return cmd
69 def construct(self: Selector,
70 target_file: Path | str,
71 performance_data: PerformanceDataFrame,
72 feature_data: FeatureDataFrame,
73 objective: SparkleObjective,
74 runtime_cutoff: int | float | str = None,
75 wallclock_limit: int | float | str = None,
76 run_on: Runner = Runner.SLURM,
77 sbatch_options: list[str] = None,
78 base_dir: Path = Path()) -> Run:
79 """Construct the Selector.
81 Args:
82 target_file: Path to the file to save the Selector to.
83 performance_data: Path to the performance data csv.
84 feature_data: Path to the feature data csv.
85 objective: The objective to optimize for selection.
86 runtime_cutoff: Cutoff for the runtime in seconds.
87 wallclock_limit: Cutoff for the wallclock time in seconds.
88 run_on: Which runner to use. Defaults to slurm.
89 sbatch_options: Additional options to pass to sbatch.
90 base_dir: The base directory to run the Selector in.
92 Returns:
93 Path to the constructed Selector.
94 """
95 if isinstance(target_file, str):
96 target_file = self.raw_output_directory / target_file
97 # Convert the dataframes to Selector Format
98 performance_csv = performance_data.to_autofolio(objective=objective,
99 target=target_file.parent)
100 feature_csv = feature_data.to_autofolio(target_file.parent)
101 cmd = self.build_construction_cmd(target_file,
102 performance_csv,
103 feature_csv,
104 objective,
105 runtime_cutoff,
106 wallclock_limit)
108 cmd_str = " ".join([str(c) for c in cmd])
109 construct = rrr.add_to_queue(
110 runner=run_on,
111 cmd=[cmd_str],
112 name="construct_selector",
113 base_dir=base_dir,
114 stdout=Path("normal.log"),
115 stderr=Path("error.log"),
116 sbatch_options=sbatch_options)
117 if run_on == Runner.LOCAL:
118 construct.wait()
119 if not target_file.is_file():
120 print(f"Selector construction of {self.name} failed!")
122 return construct
124 def build_cmd(self: Selector,
125 selector_path: Path,
126 feature_vector: list | str) -> list[str | Path]:
127 """Builds the commandline call string for running the Selector."""
128 if isinstance(feature_vector, list):
129 feature_vector = " ".join(map(str, feature_vector))
131 return ["python3", self.selector_builder_path,
132 "--load", selector_path,
133 "--feature_vec", feature_vector]
135 def run(self: Selector,
136 selector_path: Path,
137 feature_vector: list | str) -> list:
138 """Run the Selector, returning the prediction schedule upon success."""
139 cmd = self.build_cmd(selector_path, feature_vector)
140 run = subprocess.run(cmd, capture_output=True)
141 if run.returncode != 0:
142 print(f"Selector run of {self.name} failed! Error:\n"
143 f"{run.stderr.decode()}")
144 return None
145 # Process the prediction schedule from the output
146 schedule = Selector.process_predict_schedule_output(run.stdout.decode())
147 if schedule is None:
148 print(f"Error getting predict schedule! Selector {self.name} output:\n"
149 f"{run.stderr.decode()}")
150 return schedule
152 @staticmethod
153 def process_predict_schedule_output(output: str) -> list:
154 """Return the predicted algorithm schedule as a list."""
155 prefix_string = "Selected Schedule [(algorithm, budget)]: "
156 predict_schedule = ""
157 predict_schedule_lines = output.splitlines()
158 for line in predict_schedule_lines:
159 if line.strip().startswith(prefix_string):
160 predict_schedule = line.strip()
161 break
162 if predict_schedule == "":
163 return None
164 predict_schedule_string = predict_schedule[len(prefix_string):]
165 return ast.literal_eval(predict_schedule_string)