Set Hammersley as default sampling

deepxde/data/fpde.py

@@ -74,7 +74,7 @@ def __init__(
         meshtype="dynamic",
         num_domain=0,
         num_boundary=0,
-        train_distribution="Sobol",
+        train_distribution="Hammersley",
         anchors=None,
         solution=None,
         num_test=None,
@@ -217,7 +217,7 @@ def __init__(
         num_domain=0,
         num_boundary=0,
         num_initial=0,
-        train_distribution="Sobol",
+        train_distribution="Hammersley",
         anchors=None,
         solution=None,
         num_test=None,
@@ -312,7 +312,9 @@ def train_points(self):
             if self.train_distribution == "uniform":
                 tmp = self.geom.uniform_initial_points(self.num_initial)
             else:
-                tmp = self.geom.random_initial_points(self.num_initial, random="Sobol")
+                tmp = self.geom.random_initial_points(
+                    self.num_initial, random=self.train_distribution
+                )
             X = np.vstack((tmp, X))
         return X
 

deepxde/data/func_constraint.py

@@ -46,7 +46,9 @@ def train_next_batch(self, batch_size=None):
         if self.dist_train == "log uniform":
             self.train_x = self.geom.log_uniform_points(self.num_train, False)
         elif self.dist_train == "random":
-            self.train_x = self.geom.random_points(self.num_train, "Sobol")
+            self.train_x = self.geom.random_points(
+                self.num_train, random=self.dist_train
+            )
         else:
             self.train_x = self.geom.uniform_points(self.num_train, False)
         if self.anchors is not None:

deepxde/data/ide.py

@@ -25,7 +25,7 @@ def __init__(
         kernel=None,
         num_domain=0,
         num_boundary=0,
-        train_distribution="Sobol",
+        train_distribution="Hammersley",
         anchors=None,
         solution=None,
         num_test=None,

deepxde/data/mf.py

@@ -41,8 +41,8 @@ def train_next_batch(self, batch_size=None):
         else:
             self.X_train = np.vstack(
                 (
-                    self.geom.random_points(self.num_lo, "Sobol"),
-                    self.geom.random_points(self.num_hi, "Sobol"),
+                    self.geom.random_points(self.num_lo, random=self.dist_train),
+                    self.geom.random_points(self.num_hi, random=self.dist_train),
                 )
             )
         y_lo_train = self.func_lo(self.X_train)

deepxde/data/pde.py

@@ -78,7 +78,7 @@ def __init__(
         bcs,
         num_domain=0,
         num_boundary=0,
-        train_distribution="Sobol",
+        train_distribution="Hammersley",
         anchors=None,
         exclusions=None,
         solution=None,
@@ -91,19 +91,6 @@ def __init__(
 
         self.num_domain = num_domain
         self.num_boundary = num_boundary
-        if train_distribution not in [
-            "uniform",
-            "pseudo",
-            "LHS",
-            "Halton",
-            "Hammersley",
-            "Sobol",
-        ]:
-            raise ValueError(
-                "train_distribution == {} is not available choices.".format(
-                    train_distribution
-                )
-            )
         self.train_distribution = train_distribution
         self.anchors = None if anchors is None else anchors.astype(config.real(np))
         self.exclusions = exclusions
@@ -289,7 +276,7 @@ def __init__(
         num_domain=0,
         num_boundary=0,
         num_initial=0,
-        train_distribution="Sobol",
+        train_distribution="Hammersley",
         anchors=None,
         exclusions=None,
         solution=None,

deepxde/geometry/sampler.py

@@ -1,3 +1,5 @@
+__all__ = ["sample"]
+
 from distutils.version import LooseVersion
 
 import numpy as np
@@ -7,7 +9,7 @@
 
 
 def sample(n_samples, dimension, sampler="pseudo"):
-    """Generate random or quasirandom samples in [0, 1]^dimension.
+    """Generate pseudorandom or quasirandom samples in [0, 1]^dimension.
 
     Args:
         n_samples (int): The number of samples.
@@ -17,13 +19,13 @@ def sample(n_samples, dimension, sampler="pseudo"):
             sequence), or "Sobol" (Sobol sequence).
     """
     if sampler == "pseudo":
-        return pseudo(n_samples, dimension)
+        return pseudorandom(n_samples, dimension)
     if sampler in ["LHS", "Halton", "Hammersley", "Sobol"]:
         return quasirandom(n_samples, dimension, sampler)
-    raise ValueError("f{sampler} sampler is not available.")
+    raise ValueError("f{sampler} sampling is not available.")
 
 
-def pseudo(n_samples, dimension):
+def pseudorandom(n_samples, dimension):
     """Pseudo random."""
     # If random seed is set, then the rng based code always returns the same random
     # number, which may not be what we expect.