Implement custom sampling logic

Author: FilippoOlivo

pina/domain/cartesian.py

@@ -160,10 +160,10 @@ def _1d_sampler(n, mode, variables):
                     pts_variable.labels = [variable]
 
                     tmp.append(pts_variable)
-
-            result = tmp[0]
-            for i in tmp[1:]:
-                result = result.append(i, mode="cross")
+            if tmp:
+                result = tmp[0]
+                for i in tmp[1:]:
+                    result = result.append(i, mode="cross")
 
             for variable in variables:
                 if variable in self.fixed_.keys():
@@ -242,6 +242,8 @@ def _single_points_sample(n, variables):
 
         if self.fixed_ and (not self.range_):
             return _single_points_sample(n, variables)
+        if isinstance(variables, str) and variables in self.fixed_.keys():
+            return _single_points_sample(n, variables)
 
         if mode in ["grid", "chebyshev"]:
             return _1d_sampler(n, mode, variables).extract(variables)

pina/problem/abstract_problem.py

@@ -2,11 +2,12 @@
 
 from abc import ABCMeta, abstractmethod
 from ..utils import check_consistency
-from ..domain import DomainInterface
+from ..domain import DomainInterface, CartesianDomain
 from ..condition.domain_equation_condition import DomainEquationCondition
 from ..condition import InputPointsEquationCondition
 from copy import deepcopy
-from pina import LabelTensor
+from .. import LabelTensor
+from ..utils import merge_tensors
 
 
 class AbstractProblem(metaclass=ABCMeta):
@@ -21,7 +22,7 @@ class AbstractProblem(metaclass=ABCMeta):
 
     def __init__(self):
 
-        self.discretised_domains = {}
+        self._discretised_domains = {}
         # create collector to manage problem data
 
         # create hook conditions <-> problems
@@ -53,6 +54,10 @@ def batching_dimension(self):
     def batching_dimension(self, value):
         self._batching_dimension = value
 
+    @property
+    def discretised_domains(self):
+        return self._discretised_domains
+
     # TODO this should be erase when dataloading will interface collector,
     # kept only for back compatibility
     @property
@@ -62,7 +67,7 @@ def input_pts(self):
             if hasattr(cond, "input_points"):
                 to_return[cond_name] = cond.input_points
             elif hasattr(cond, "domain"):
-                to_return[cond_name] = self.discretised_domains[cond.domain]
+                to_return[cond_name] = self._discretised_domains[cond.domain]
         return to_return
 
     def __deepcopy__(self, memo):
@@ -139,9 +144,10 @@ def conditions(self):
         return self.conditions
 
     def discretise_domain(self,
-                          n,
+                          n=None,
                           mode="random",
-                          domains="all"):
+                          domains="all",
+                          sample_rules=None):
         """
         Generate a set of points to span the `Location` of all the conditions of
         the problem.
@@ -153,6 +159,8 @@ def discretise_domain(self,
             Available modes include: random sampling, ``random``;
             latin hypercube sampling, ``latin`` or ``lh``;
             chebyshev sampling, ``chebyshev``; grid sampling ``grid``.
+        :param variables: variable(s) to sample, defaults to 'all'.
+        :type variables: str | list[str]
         :param domains: problem's domain from where to sample, defaults to 'all'.
         :type domains: str | list[str]
 
@@ -170,24 +178,55 @@ def discretise_domain(self,
         """
 
         # check consistecy n, mode, variables, locations
-        check_consistency(n, int)
-        check_consistency(mode, str)
+        if sample_rules is not None:
+            check_consistency(sample_rules, dict)
+        if mode is not None:
+            check_consistency(mode, str)
         check_consistency(domains, (list, str))
 
-        # check correct sampling mode
-        # if mode not in DomainInterface.available_sampling_modes:
-        #     raise TypeError(f"mode {mode} not valid.")
-
         # check correct location
         if domains == "all":
             domains = self.domains.keys()
         elif not isinstance(domains, (list)):
             domains = [domains]
+        if n is not None and sample_rules is None:
+            self._apply_default_discretization(n, mode, domains)
+        if n is None and sample_rules is not None:
+            self._apply_custom_discretization(sample_rules, domains)
+        elif n is not None and sample_rules is not None:
+            raise RuntimeError(
+                "You can't specify both n and sample_rules at the same time."
+            )
+        elif n is None and sample_rules is None:
+            raise RuntimeError(
+                "You have to specify either n or sample_rules."
+            )
 
+    def _apply_default_discretization(self, n, mode, domains):
         for domain in domains:
             self.discretised_domains[domain] = (
-                self.domains[domain].sample(n, mode)
+                self.domains[domain].sample(n, mode).sort_labels()
+            )
+
+    def _apply_custom_discretization(self, sample_rules, domains):
+        if sorted(list(sample_rules.keys())) != sorted(self.input_variables):
+            raise RuntimeError(
+                "The keys of the sample_rules dictionary must be the same as "
+                "the input variables."
             )
+        for domain in domains:
+            if not isinstance(self.domains[domain], CartesianDomain):
+                raise RuntimeError(
+                    "Custom discretisation can be applied only on Cartesian "
+                    "domains")
+            discretised_tensor = []
+            for var, rules in sample_rules.items():
+                n, mode = rules['n'], rules['mode']
+                points = self.domains[domain].sample(n, mode, var)
+                discretised_tensor.append(points)
+
+            self.discretised_domains[domain] = merge_tensors(
+                discretised_tensor).sort_labels()
 
     def add_points(self, new_points_dict):
         """

tests/test_collector.py

@@ -11,22 +11,23 @@
 from pina.collector import Collector
 
 
-# def test_supervised_tensor_collector():
-#     class SupervisedProblem(AbstractProblem):
-#         output_variables = None
-#         conditions = {
-#             'data1' : Condition(input_points=torch.rand((10,2)),
-#                                 output_points=torch.rand((10,2))),
-#             'data2' : Condition(input_points=torch.rand((20,2)),
-#                                 output_points=torch.rand((20,2))),
-#             'data3' : Condition(input_points=torch.rand((30,2)),
-#                                 output_points=torch.rand((30,2))),
-#         }
-#     problem = SupervisedProblem()
-#     collector = Collector(problem)
-#     for v in collector.conditions_name.values():
-#         assert v in problem.conditions.keys()
-#     assert all(collector._is_conditions_ready.values())
+def test_supervised_tensor_collector():
+    class SupervisedProblem(AbstractProblem):
+        output_variables = None
+        conditions = {
+            'data1': Condition(input_points=torch.rand((10, 2)),
+                               output_points=torch.rand((10, 2))),
+            'data2': Condition(input_points=torch.rand((20, 2)),
+                               output_points=torch.rand((20, 2))),
+            'data3': Condition(input_points=torch.rand((30, 2)),
+                               output_points=torch.rand((30, 2))),
+        }
+
+    problem = SupervisedProblem()
+    collector = Collector(problem)
+    for v in collector.conditions_name.values():
+        assert v in problem.conditions.keys()
+
 
 def test_pinn_collector():
     def laplace_equation(input_, output_):
@@ -81,7 +82,7 @@ class Poisson(SpatialProblem):
         def poisson_sol(self, pts):
             return -(torch.sin(pts.extract(['x']) * torch.pi) *
                      torch.sin(pts.extract(['y']) * torch.pi)) / (
-                        2 * torch.pi ** 2)
+                    2 * torch.pi ** 2)
 
         truth_solution = poisson_sol
 

tests/test_problem.py

@@ -2,6 +2,8 @@
 import pytest
 from pina.problem.zoo import Poisson2DSquareProblem as Poisson
 from pina import LabelTensor
+from pina.domain import Union
+from pina.domain import CartesianDomain
 
 
 def test_discretise_domain():
@@ -29,18 +31,6 @@ def test_discretise_domain():
     poisson_problem.discretise_domain(n)
 
 
-'''
-def test_sampling_few_variables():
-    n = 10
-    poisson_problem = Poisson()
-    poisson_problem.discretise_domain(n,
-                                      'grid',
-                                      domains=['D'],
-                                      variables=['x'])
-    assert poisson_problem.discretised_domains['D'].shape[1] == 1
-'''
-
-
 def test_variables_correct_order_sampling():
     n = 10
     poisson_problem = Poisson()
@@ -66,3 +56,39 @@ def test_add_points():
                          new_pts.extract('x'))
     assert torch.isclose(poisson_problem.discretised_domains['D'].extract('y'),
                          new_pts.extract('y'))
+
+@pytest.mark.parametrize(
+    "mode",
+    [
+        'random',
+        'grid'
+    ]
+)
+def test_custom_sampling_logic(mode):
+    poisson_problem = Poisson()
+    sampling_rules = {
+        'x': {'n': 100, 'mode': mode},
+        'y': {'n': 50, 'mode': mode}
+    }
+    poisson_problem.discretise_domain(sample_rules=sampling_rules)
+    for domain in ['g1', 'g2', 'g3', 'g4']:
+        assert poisson_problem.discretised_domains[domain].shape[0] == 100 * 50
+        assert poisson_problem.discretised_domains[domain].labels == ['x', 'y']
+
+@pytest.mark.parametrize(
+    "mode",
+    [
+        'random',
+        'grid'
+    ]
+)
+def test_wrong_custom_sampling_logic(mode):
+    d2 = CartesianDomain({'x': [1,2], 'y': [0,1] })
+    poisson_problem = Poisson()
+    poisson_problem.domains['D'] = Union([poisson_problem.domains['D'], d2])
+    sampling_rules = {
+        'x': {'n': 100, 'mode': mode},
+        'y': {'n': 50, 'mode': mode}
+    }
+    with pytest.raises(RuntimeError):
+        poisson_problem.discretise_domain(sample_rules=sampling_rules)