fixes to match new TFP API for mcmc sampling

greta-dev · Nov 11, 2018 · 03ec782 · 03ec782
1 parent a4c63a3
commit 03ec782
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 4 deletions.
diff --git a/DESCRIPTION b/DESCRIPTION
@@ -1,7 +1,7 @@
 Package: greta
 Type: Package
 Title: Simple and Scalable Statistical Modelling in R
-Version: 0.3.0
+Version: 0.3.0.9001
 Date: 2018-10-30
 Authors@R: c(
   person("Nick", "Golding", role = c("aut", "cre"),

diff --git a/NEWS.md b/NEWS.md
@@ -1,3 +1,8 @@
+# greta 0.3.0.9001
+
+* mcmc now works with TensorFlow Probability version 0.5.0 (#248)
+
+
 # greta 0.3.0
 
 This is a very large update which adds a number of features and major speed improvements. We now depend on the TensorFlow Probability Python package, and use functionality in that package wherever possible. Sampling a simple model now takes ~10s, rather than ~2m (>10x speedup).

diff --git a/R/inference_class.R b/R/inference_class.R
@@ -629,8 +629,9 @@ sampler <- R6Class(
           num_results = sampler_burst_length %/% sampler_thin,
           current_state = free_state,
           kernel = sampler_kernel,
-          num_burnin_steps = 0L,
-          num_steps_between_results = sampler_thin)
+          num_burnin_steps = tf$constant(0L, dtype = tf$int64),
+          num_steps_between_results = tf$cast(sampler_thin, tf$int64),
+          parallel_iterations = 1L)
       )
 
     },
@@ -763,7 +764,7 @@ hmc_sampler <- R6Class(
 
       # tensors for sampler parameters
       dag$tf_run(hmc_epsilon <- tf$placeholder(dtype = tf_float()))
-      dag$tf_run(hmc_L <- tf$placeholder(dtype = tf$int32))
+      dag$tf_run(hmc_L <- tf$placeholder(dtype = tf$int64))
 
       # need to pass in the value for this placeholder as a matrix (shape(n, 1))
       dag$tf_run(