ANL-CEEESA/MIPLearn
1
0
Fork 0
mirror of https://github.com/ANL-CEEESA/MIPLearn.git synced 2025-12-06 09:28:51 -06:00
Code Issues Packages Projects Releases Wiki Activity

Primal: reactivate before_solve_mip

Browse Source
This commit is contained in:
Alinson S. Xavier
2021-03-31 12:07:58 -05:00
parent fe7bad885c
commit db2f426140
7 changed files with 133 additions and 102 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
miplearn/benchmark.py
View File

@@ -10,7 +10,8 @@ import pandas as pd
from miplearn.instance import Instance
from miplearn.solvers.learning import LearningSolver
from miplearn.types import LearningSolveStats
logger = logging.getLogger(__name__)
class BenchmarkRunner:
@@ -110,6 +111,7 @@ class BenchmarkRunner:
"""
for (solver_name, solver) in self.solvers.items():
logger.debug(f"Fitting {solver_name}...")
solver.fit(instances)
def _silence_miplearn_logger(self) -> None:

11
miplearn/classifiers/__init__.py
View File

@@ -34,7 +34,11 @@ class Classifier(ABC):
"""
assert isinstance(x_train, np.ndarray)
assert isinstance(y_train, np.ndarray)
assert x_train.dtype in [np.float16, np.float32, np.float64]
assert x_train.dtype in [
np.float16,
np.float32,
np.float64,
], f"x_train.dtype shoule be float. Found {x_train.dtype} instead."
assert y_train.dtype == np.bool8
assert len(x_train.shape) == 2
assert len(y_train.shape) == 2
@@ -67,7 +71,10 @@ class Classifier(ABC):
assert isinstance(x_test, np.ndarray)
assert len(x_test.shape) == 2
(n_samples, n_features_x) = x_test.shape
assert n_features_x == self.n_features
assert n_features_x == self.n_features, (
f"Test and training data have different number of "
f"features: {n_features_x} != {self.n_features}"
)
return np.ndarray([])

78
miplearn/components/primal.py
View File

@@ -24,11 +24,9 @@ from miplearn.classifiers.adaptive import AdaptiveClassifier
from miplearn.classifiers.threshold import MinPrecisionThreshold, Threshold
from miplearn.components import classifier_evaluation_dict
from miplearn.components.component import Component
from miplearn.extractors import InstanceIterator
from miplearn.instance import Instance
from miplearn.types import (
TrainingSample,
VarIndex,
Solution,
LearningSolveStats,
Features,
@@ -70,30 +68,33 @@ class PrimalSolutionComponent(Component):
self._n_one = 0
def before_solve_mip(self, solver, instance, model):
pass
# if len(self.thresholds) > 0:
# logger.info("Predicting primal solution...")
# solution = self.predict(instance)
#
# # Collect prediction statistics
# self._n_free = 0
# self._n_zero = 0
# self._n_one = 0
# for (var, var_dict) in solution.items():
# for (idx, value) in var_dict.items():
# if value is None:
# self._n_free += 1
# else:
# if value < 0.5:
# self._n_zero += 1
# else:
# self._n_one += 1
#
# # Provide solution to the solver
# if self.mode == "heuristic":
# solver.internal_solver.fix(solution)
# else:
# solver.internal_solver.set_warm_start(solution)
if len(self.thresholds) > 0:
logger.info("Predicting primal solution...")
solution = self.predict(instance.features, instance.training_data[-1])
# Collect prediction statistics
self._n_free = 0
self._n_zero = 0
self._n_one = 0
for (var, var_dict) in solution.items():
for (idx, value) in var_dict.items():
if value is None:
self._n_free += 1
else:
if value < 0.5:
self._n_zero += 1
else:
self._n_one += 1
logger.info(
f"Predicted: {self._n_free} free, {self._n_zero} fix-zero, "
f"{self._n_one} fix-one"
)
# Provide solution to the solver
if self.mode == "heuristic":
solver.internal_solver.fix(solution)
else:
solver.internal_solver.set_warm_start(solution)
def after_solve_mip(
self,
@@ -120,27 +121,29 @@ class PrimalSolutionComponent(Component):
self.classifiers[category] = clf
self.thresholds[category] = thr
def predict(self, instance: Instance) -> Solution:
assert len(instance.training_data) > 0
sample = instance.training_data[-1]
def predict(
self,
features: Features,
sample: TrainingSample,
) -> Solution:
# Initialize empty solution
solution: Solution = {}
for (var_name, var_dict) in instance.features["Variables"].items():
for (var_name, var_dict) in features["Variables"].items():
solution[var_name] = {}
for idx in var_dict.keys():
solution[var_name][idx] = None
# Compute y_pred
x = self.x_sample(instance.features, sample)
x = self.x_sample(features, sample)
y_pred = {}
for category in x.keys():
assert category in self.classifiers, (
f"Classifier for category {category} has not been trained. "
f"Please call component.fit before component.predict."
)
proba = self.classifiers[category].predict_proba(x[category])
thr = self.thresholds[category].predict(x[category])
xc = np.array(x[category])
proba = self.classifiers[category].predict_proba(xc)
thr = self.thresholds[category].predict(xc)
y_pred[category] = np.vstack(
[
proba[:, 0] > thr[0],
@@ -150,7 +153,7 @@ class PrimalSolutionComponent(Component):
# Convert y_pred into solution
category_offset: Dict[Hashable, int] = {cat: 0 for cat in x.keys()}
for (var_name, var_dict) in instance.features["Variables"].items():
for (var_name, var_dict) in features["Variables"].items():
for (idx, var_features) in var_dict.items():
category = var_features["Category"]
offset = category_offset[category]
@@ -250,8 +253,9 @@ class PrimalSolutionComponent(Component):
if category not in x.keys():
x[category] = []
y[category] = []
f = var_features["User features"]
assert f is not None
f: List[float] = []
assert var_features["User features"] is not None
f += var_features["User features"]
if "LP solution" in sample and sample["LP solution"] is not None:
lp_value = sample["LP solution"][var_name][idx]
if lp_value is not None:

4
miplearn/problems/tsp.py
View File

@@ -157,10 +157,10 @@ class TravelingSalesmanInstance(Instance):
return model
def get_instance_features(self):
return [1]
return [0.0]
def get_variable_features(self, var_name, index):
return [1]
return [0.0]
def get_variable_category(self, var_name, index):
return index

3
miplearn/solvers/learning.py
View File

@@ -373,6 +373,9 @@ class LearningSolver:
The list is the same you would obtain by calling
`[solver.solve(p) for p in instances]`
"""
if n_jobs == 1:
return [self.solve(p) for p in instances]
else:
self.internal_solver = None
self._silence_miplearn_logger()
_GLOBAL[0].solver = self

50
tests/components/test_primal.py
View File

@@ -2,7 +2,6 @@
# Copyright (C) 2020, UChicago Argonne, LLC. All rights reserved.
# Released under the modified BSD license. See COPYING.md for more details.
from typing import cast
from unittest.mock import Mock
import numpy as np
@@ -11,7 +10,6 @@ from numpy.testing import assert_array_equal
from miplearn import Classifier
from miplearn.classifiers.threshold import Threshold
from miplearn.components.primal import PrimalSolutionComponent
from miplearn.instance import Instance
from miplearn.types import TrainingSample, Features
@@ -142,8 +140,7 @@ def test_predict() -> None:
)
thr = Mock(spec=Threshold)
thr.predict = Mock(return_value=[0.75, 0.75])
instance = cast(Instance, Mock(spec=Instance))
instance.features = {
features: Features = {
"Variables": {
"x": {
0: {
@@ -161,8 +158,7 @@ def test_predict() -> None:
}
}
}
instance.training_data = [
{
sample: TrainingSample = {
"LP solution": {
"x": {
0: 0.1,
@@ -171,23 +167,14 @@ def test_predict() -> None:
}
}
}
]
x = {
"default": np.array(
[
[0.0, 0.0, 0.1],
[0.0, 2.0, 0.5],
[2.0, 0.0, 0.9],
]
)
}
x = PrimalSolutionComponent.x_sample(features, sample)
comp = PrimalSolutionComponent()
comp.classifiers = {"default": clf}
comp.thresholds = {"default": thr}
solution_actual = comp.predict(instance)
solution_actual = comp.predict(features, sample)
clf.predict_proba.assert_called_once()
thr.predict.assert_called_once()
assert_array_equal(x["default"], clf.predict_proba.call_args[0][0])
thr.predict.assert_called_once()
assert_array_equal(x["default"], thr.predict.call_args[0][0])
assert solution_actual == {
"x": {
@@ -196,3 +183,30 @@ def test_predict() -> None:
2: 1.0,
}
}
def test_fit_xy():
comp = PrimalSolutionComponent(
classifier=lambda: Mock(spec=Classifier),
threshold=lambda: Mock(spec=Threshold),
)
x = {
"type-a": np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]),
"type-b": np.array([[7.0, 8.0, 9.0]]),
}
y = {
"type-a": np.array([[True, False], [False, True]]),
"type-b": np.array([[True, False]]),
}
comp.fit_xy(x, y)
for category in ["type-a", "type-b"]:
assert category in comp.classifiers
assert category in comp.thresholds
clf = comp.classifiers[category]
clf.fit.assert_called_once()
assert_array_equal(x[category], clf.fit.call_args[0][0])
assert_array_equal(y[category], clf.fit.call_args[0][1])
thr = comp.thresholds[category]
thr.fit.assert_called_once()
assert_array_equal(x[category], thr.fit.call_args[0][1])
assert_array_equal(y[category], thr.fit.call_args[0][2])

9
tests/test_benchmark.py
View File

@@ -12,23 +12,24 @@ from miplearn.solvers.learning import LearningSolver
def test_benchmark():
for n_jobs in [1, 4]:
# Generate training and test instances
generator = MaxWeightStableSetGenerator(n=randint(low=25, high=26))
train_instances = generator.generate(5)
test_instances = generator.generate(3)
# Training phase...
# Solve training instances
training_solver = LearningSolver()
training_solver.parallel_solve(train_instances, n_jobs=10)
training_solver.parallel_solve(train_instances, n_jobs=n_jobs)
# Test phase...
# Benchmark
test_solvers = {
"Strategy A": LearningSolver(),
"Strategy B": LearningSolver(),
}
benchmark = BenchmarkRunner(test_solvers)
benchmark.fit(train_instances)
benchmark.parallel_solve(test_instances, n_jobs=2, n_trials=2)
benchmark.parallel_solve(test_instances, n_jobs=n_jobs, n_trials=2)
assert benchmark.results.values.shape == (12, 18)
benchmark.write_csv("/tmp/benchmark.csv")