LearningSolver: add method for solving training instances in parallel

README.md

@@ -101,6 +101,26 @@ for instance in more_instances:
 In addition to storing the training data, `solver.save` also serializes and stores all trained ML models themselves, so it is not necessary to call `solver.fit`.
 
 
+### Solving training instances in parallel
+
+In many situations, training instances can be solved in parallel to accelerate the training process. `LearningSolver` provides the method `parallel_solve(instances)` to easily achieve this. After all instances have been solved, the ML models can be trained and saved to disk as usual, as the next example illustrates:
+
+```python
+from miplearn import LearningSolver
+
+# Training phase...
+solver = LearningSolver(...) # training solver parameters
+solver.parallel_solve(training_instances, n_jobs=4)
+solver.fit()
+solver.save("/tmp/data.bin")
+
+# Test phase...
+solver = LearningSolver(...) # test solver parameters
+solver.load("/tmp/data.bin")
+solver.solve(test_instance)
+```
+
+
 Current Limitations
 -------------------
 

miplearn/solvers.py

@@ -6,8 +6,12 @@ from .transformers import PerVariableTransformer
 from .warmstart import KnnWarmStartPredictor
 import pyomo.environ as pe
 import numpy as np
-from copy import deepcopy
+from copy import copy, deepcopy
 import pickle
+from tqdm import tqdm
+from joblib import Parallel, delayed
+import multiprocessing
+
 
 
 class LearningSolver:
@@ -79,6 +83,29 @@ class LearningSolver:
             else:
                 self.y_train[category] = np.vstack([self.y_train[category], y])
                 
+    def parallel_solve(self, instances, n_jobs=4):
+        def _process(instance):
+            solver = copy(self)
+            solver.solve(instance)
+            return solver.x_train, solver.y_train
+
+        def _merge(results):
+            categories = results[0][0].keys()
+            x_entries = [np.vstack([r[0][c] for r in results]) for c in categories]
+            y_entries = [np.vstack([r[1][c] for r in results]) for c in categories]
+            x_train = dict(zip(categories, x_entries))
+            y_train = dict(zip(categories, y_entries))
+            return x_train, y_train
+
+        results = Parallel(n_jobs=n_jobs)(
+            delayed(_process)(i)
+            for i in tqdm(instances)
+        )
+        
+        x_train, y_train = _merge(results)
+        self.x_train = x_train
+        self.y_train = y_train
+
     def fit(self, x_train_dict=None, y_train_dict=None):
         if x_train_dict is None:
             x_train_dict = self.x_train

miplearn/tests/test_solver.py

@@ -4,6 +4,7 @@
 
 from miplearn import LearningSolver
 from miplearn.problems.knapsack import KnapsackInstance2
+import numpy as np
 
 
 def test_solver():
@@ -31,3 +32,12 @@ def test_solve_save_load():
     assert len(solver.x_train) == prev_x_train_len
     assert len(solver.y_train) == prev_y_train_len
 
+def test_parallel_solve():
+    instances = [KnapsackInstance2(weights=np.random.rand(5),
+                                   prices=np.random.rand(5),
+                                   capacity=3.0)
+                 for _ in range(10)]
+    solver = LearningSolver()
+    solver.parallel_solve(instances, n_jobs=2)
+    assert len(solver.x_train[0]) == 10
+    assert len(solver.y_train[0]) == 10

requirements.txt

@@ -3,3 +3,4 @@ numpy
 pytest
 sklearn
 networkx
+tqdm

setup.py

@@ -7,5 +7,11 @@ setup(
     author='Alinson S. Xavier',
     author_email='axavier@anl.gov',
     packages=['miplearn'],
-   install_requires=['pyomo', 'numpy', 'sklearn', 'networkx'],
+    install_requires=[
+        'pyomo',
+        'numpy',
+        'sklearn',
+        'networkx',
+        'tqdm',
+    ],
 )