Bump version

cascades/__init__.py

@@ -16,7 +16,7 @@
 
 # A new PyPI release will be pushed everytime `__version__` is increased
 # When changing this, also update the CHANGELOG.md
-__version__ = '0.3.2'
+__version__ = '0.4.0'
 
 from cascades._src.distributions.base import UniformCategorical
 from cascades._src.distributions.gpt import GPT

cascades/_src/distributions/base.py

@@ -20,9 +20,12 @@
 import math
 from typing import Any, Optional, Tuple, Union
 
+import cachetools
 import jax
 import jax.numpy as jnp
+import numpy as np
 from numpyro import distributions as np_dists
+import shortuuid
 
 DEFAULT_TIMEOUT = 60
 
@@ -234,3 +237,64 @@ def log_prob(self, value):
 
   def support(self):
     return self.options
+
+
+# Higher order distributons
+
+
+def _rng_hash(rng):
+  """Hash an rng key."""
+  if isinstance(rng, int):
+    return hash(rng)
+  else:
+    return hash(tuple(np.asarray(rng)))
+
+
+def _mem_sample_key(self, rng):
+  """Cache key for Mem distribution."""
+  rng = _rng_hash(rng)
+  h = hash((rng, self))
+  return h
+
+
+def _mem_score_key(self, value):
+  """Cache key for Mem distribution."""
+  h = hash((value, self))
+  return h
+
+
+# TODO(ddohan): Consider sharing cache across instances.
+@dataclasses.dataclass(frozen=True, eq=True)
+class Mem(Distribution):
+  """Add a cache to a distribution so that repeated calls are memoized."""
+  dist: Optional[Distribution] = dataclasses.field(repr=True, default=None)
+
+  # Mem should not be equal unless they are really the same
+  # generate a unique UID to ensure this property.
+  uid: str = dataclasses.field(
+      repr=True, default_factory=lambda: shortuuid.uuid()[:8])
+
+  @cachetools.cached(cache=cachetools.LRUCache(maxsize=100_000),
+                     key=_mem_sample_key)
+  def sample(self, rng):
+    return self.dist.sample(rng=rng)
+
+  @cachetools.cached(cache=cachetools.LRUCache(maxsize=100_000),
+                     key=_mem_score_key)
+  def score(self, value):
+    return self.dist.score(value)
+
+
+@dataclasses.dataclass(frozen=True, eq=True)
+class Lambda(Distribution):
+  """Wrap a function as distribution."""
+  fn: Any = dataclasses.field(
+      default=None, hash=None)  # TODO(ddohan): Add type of callable.
+
+  def sample(self, rng):
+    del rng
+    value = self.fn()
+    return RandomSample(value=value, dist=self, log_p=None)
+
+  def score(self):
+    raise NotImplementedError('Scoring from Lambda is not available.')

cascades/_src/distributions/base_test.py

@@ -0,0 +1,74 @@
+# Copyright 2022 The cascades Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for base distributions."""
+
+import dataclasses
+import random
+
+from absl.testing import absltest
+from cascades._src.distributions import base
+
+
+@dataclasses.dataclass(eq=True, frozen=True)
+class RandomFactor(base.Distribution):
+  """Randomized likelihood for testing purposes."""
+
+  def sample(self, rng):
+    del rng
+    return base.RandomSample(value=None, log_p=self.score(None))
+
+  def score(self, value):
+    del value
+    return random.randint(0, 100_000_000)
+
+
+class BaseTest(absltest.TestCase):
+
+  def test_lambda(self):
+    fn = lambda: 5
+    dist = base.Lambda(fn=fn)
+    sample = dist.sample(0)
+    self.assertEqual(5, sample.value)
+
+  def test_mem_lambda(self):
+    """Test memoizing a lambda distribution."""
+    fn = lambda: random.randint(0, 100_000_000)
+    dist = base.Lambda(fn=fn)
+    dist = base.Mem(dist=dist)
+    v1 = dist.sample(0).value
+    v2 = dist.sample(0).value
+    v3 = dist.sample(1).value
+    self.assertEqual(v1, v2)
+    self.assertNotEqual(v1, v3)
+
+  def test_mem_sample_and_score(self):
+    """Test memoizing a randomized sample & score distribution."""
+    dist = RandomFactor()
+    dist = base.Mem(dist=dist)
+    v1 = dist.sample(0).score
+    v2 = dist.sample(0).score
+    v3 = dist.sample(1).score
+    self.assertEqual(v1, v2)
+    self.assertNotEqual(v1, v3)
+
+    v1 = dist.score('abc')
+    v2 = dist.score('abc')
+    v3 = dist.score('xyz')
+    self.assertEqual(v1, v2)
+    self.assertNotEqual(v1, v3)
+
+
+if __name__ == '__main__':
+  absltest.main()

pyproject.toml

@@ -14,7 +14,7 @@ classifiers = [
 keywords = []
 
 # pip dependencies of the project
-dependencies = ["jax[cpu]", "numpyro", "immutabledict", "openai"]
+dependencies = ["cachetools", "shortuuid", "jax[cpu]", "numpyro", "immutabledict", "openai"]
 
 # This is set automatically by flit using `cascades.__version__`
 dynamic = ["version"]