FunMC: Add AIS/SMC.

PiperOrigin-RevId: 492849049

spinoffs/fun_mc/fun_mc/dynamic/backend_jax/tf_on_jax.py

@@ -161,8 +161,10 @@ def _get_static_value(value):
 
 tf.newaxis = None
 
+_impl_np()(jnp.cumsum)
 _impl_np()(jnp.exp)
 _impl_np()(jnp.einsum)
+_impl_np()(jnp.floor)
 _impl_np()(jnp.float32)
 _impl_np()(jnp.float64)
 _impl_np()(jnp.int32)
@@ -179,20 +181,22 @@ def _get_static_value(value):
 _impl_np()(jnp.zeros_like)
 _impl_np()(jnp.transpose)
 _impl_np(name='fill')(jnp.full)
+_impl_np(['nn'])(jax.nn.softmax)
 _impl_np(['math'])(jnp.ceil)
 _impl_np(['math'])(jnp.log)
 _impl_np(['math'], name='mod')(jnp.mod)
 _impl_np(['math'])(jnp.sqrt)
 _impl_np(['math'], name='is_finite')(jnp.isfinite)
+_impl_np(['math'], name='is_nan')(jnp.isnan)
 _impl_np(['math'], name='pow')(jnp.power)
+_impl_np(['math'], name='reduce_all')(jnp.all)
 _impl_np(['math'], name='reduce_prod')(jnp.prod)
 _impl_np(['math'], name='reduce_variance')(jnp.var)
 _impl_np(name='abs')(jnp.abs)
 _impl_np(name='Tensor')(jnp.ndarray)
 _impl_np(name='concat')(jnp.concatenate)
 _impl_np(name='constant')(jnp.array)
 _impl_np(name='expand_dims')(jnp.expand_dims)
-_impl_np(['math'], name='reduce_all')(jnp.all)
 _impl_np(name='reduce_max')(jnp.max)
 _impl_np(name='reduce_mean')(jnp.mean)
 _impl_np(name='reduce_sum')(jnp.sum)

spinoffs/fun_mc/fun_mc/dynamic/backend_jax/util.py

@@ -20,12 +20,12 @@
 from jax import lax
 from jax import random
 from jax import tree_util
-from jax.example_libraries import stax
 import jax.numpy as jnp
 
 __all__ = [
     'assert_same_shallow_tree',
     'block_until_ready',
+    'diff',
     'flatten_tree',
     'get_shallow_tree',
     'inverse_fn',
@@ -38,6 +38,7 @@
     'random_integer',
     'random_normal',
     'random_uniform',
+    'repeat',
     'split_seed',
     'trace',
     'value_and_grad',
@@ -143,7 +144,7 @@ def body(state):
 
 def random_categorical(logits, num_samples, seed):
   """Returns a sample from a categorical distribution. `logits` must be 2D."""
-  probs = stax.softmax(logits)
+  probs = jax.nn.softmax(logits)
   cum_sum = jnp.cumsum(probs, axis=-1)
 
   eta = random.uniform(
@@ -211,6 +212,7 @@ def wrapper(i, state_untraced_traced):
       state, untraced, traced = fn(state)
       trace_arrays = map_tree(lambda a, e: a.at[i].set(e), trace_arrays, traced)
       return (state, untraced, trace_arrays)
+
     state, untraced, traced = lax.fori_loop(
         jnp.asarray(0, num_steps.dtype),
         num_steps,
@@ -250,7 +252,6 @@ def scale_by_two(x):
   assert y_extra == 3
   assert y_ldj == jnp.log(2)
   ```
-
   """
   value, (extra, ldj) = fn(args)
   return value, (extra, ldj), ldj
@@ -307,11 +308,13 @@ def block_until_ready(tensors):
   Returns:
     tensors: Tensors that are are guaranteed to be ready to materialize.
   """
+
   def _block_until_ready(tensor):
     if hasattr(tensor, 'block_until_ready'):
       return tensor.block_until_ready()
     else:
       return tensor
+
   return map_tree(_block_until_ready, tensors)
 
 
@@ -326,3 +329,13 @@ def named_call(f=None, name=None):
     return functools.partial(named_call, name=name)
 
   return jax.named_call(f, name=name)
+
+
+def diff(x, prepend=None):
+  """Like jnp.diff."""
+  return jnp.diff(x, prepend=prepend)
+
+
+def repeat(x, repeats, total_repeat_length=None):
+  """Like jnp.repeat."""
+  return jnp.repeat(x, repeats, total_repeat_length=total_repeat_length)

spinoffs/fun_mc/fun_mc/dynamic/backend_tensorflow/util.py

@@ -26,6 +26,7 @@
 __all__ = [
     'assert_same_shallow_tree',
     'block_until_ready',
+    'diff',
     'flatten_tree',
     'get_shallow_tree',
     'inverse_fn',
@@ -38,6 +39,7 @@
     'random_integer',
     'random_normal',
     'random_uniform',
+    'repeat',
     'split_seed',
     'trace',
     'value_and_ldj',
@@ -177,7 +179,9 @@ def trace(state, fn, num_steps, unroll, parallel_iterations=10):
     state, first_untraced, first_traced = fn(state)
     arrays = tf.nest.map_structure(
         lambda v: tf.TensorArray(  # pylint: disable=g-long-lambda
-            v.dtype, size=num_steps, element_shape=v.shape).write(0, v),
+            v.dtype,
+            size=num_steps,
+            element_shape=v.shape).write(0, v),
         first_traced)
     start_idx = 1
   else:
@@ -189,7 +193,10 @@ def trace(state, fn, num_steps, unroll, parallel_iterations=10):
 
     arrays = tf.nest.map_structure(
         lambda spec: tf.TensorArray(  # pylint: disable=g-long-lambda
-            spec.dtype, size=num_steps, element_shape=spec.shape), traced_spec)
+            spec.dtype,
+            size=num_steps,
+            element_shape=spec.shape),
+        traced_spec)
     first_untraced = tf.nest.map_structure(
         lambda spec: tf.zeros(spec.shape, spec.dtype), untraced_spec)
     start_idx = 0
@@ -266,7 +273,6 @@ def scale_by_two(x):
   assert y_extra == 3
   assert y_ldj == np.log(2)
   ```
-
   """
   value, (extra, ldj) = fn(args)
   return value, (extra, ldj), ldj
@@ -348,3 +354,19 @@ def wrapped(*args, **kwargs):
       return f(*args, **kwargs)
 
   return wrapped
+
+
+def diff(x, prepend=None):
+  """Like jnp.diff."""
+  if prepend is not None:
+    x = tf.concat([tf.convert_to_tensor(prepend, dtype=x.dtype)[tf.newaxis], x],
+                  0)
+  return x[1:] - x[:-1]
+
+
+def repeat(x, repeats, total_repeat_length=None):
+  """Like jnp.repeat."""
+  res = tf.repeat(x, repeats)
+  if total_repeat_length is not None:
+    res.set_shape([total_repeat_length] + [None] * (len(res.shape) - 1))
+  return res