sample_ppc bug fix (#2748)

* sample_ppc bug fix

My fix in #2725 to use all chains breaks the `sample_ppc([point]...)` and also the progress bar. These issue should be fix here now and also add test for sample_ppc from a list.

* fixe sample_ppc_w, edited release note

* Add docstring, remove space

Author: Junpeng Lao

RELEASE-NOTES.md

@@ -12,6 +12,7 @@
 - Update loo, new improved algorithm (#2730)
 - New CSG (Constant Stochastic Gradient) approximate posterior sampling
   algorithm (#2544)
+- Michael Osthege added support for population-samplers and implemented differential evolution metropolis (`DEMetropolis`).  For models with correlated dimensions that can not use gradient-based samplers, the `DEMetropolis` sampler can give higher effective sampling rates. (also see [PR#2735](https://github.com/pymc-devs/pymc3/pull/2735))
 
 ### Fixes
 
@@ -20,12 +21,9 @@
 - `sample_ppc_w` now broadcasts
 - `df_summary` function renamed to `summary`
 - Add test for `model.logp_array` and `model.bijection` (#2724) 
-- Fixed `sample_ppc` and `sample_ppc_w` to iterate all chains(#2633)
+- Fixed `sample_ppc` and `sample_ppc_w` to iterate all chains(#2633, #2748)
 - Add Bayesian R2 score (for GLMs) `stats.r2_score` (#2696) and test (#2729).
 
-### New Features
-- Michael Osthege added support for population-samplers and implemented differential evolution metropolis (`DEMetropolis`).  For models with correlated dimensions that can not use gradient-based samplers, the `DEMetropolis` sampler can give higher effective sampling rates. (also see [PR#2735](https://github.com/pymc-devs/pymc3/pull/2735))
-
 
 ## PyMC3 3.2 (October 10, 2017)
 

pymc3/sampling.py

@@ -944,7 +944,8 @@ def sample_ppc(trace, samples=None, model=None, vars=None, size=None,
     Parameters
     ----------
     trace : backend, list, or MultiTrace
-        Trace generated from MCMC sampling.
+        Trace generated from MCMC sampling. Or a list containing dicts from
+        find_MAP() or points
     samples : int
         Number of posterior predictive samples to generate. Defaults to the
         length of `trace`
@@ -971,7 +972,10 @@ def sample_ppc(trace, samples=None, model=None, vars=None, size=None,
         posterior predictive samples.
     """
     len_trace = len(trace)
-    nchain = trace.nchains
+    try:
+        nchain = trace.nchains
+    except AttributeError:
+        nchain = 1
 
     if samples is None:
         samples = len(trace)
@@ -984,14 +988,19 @@ def sample_ppc(trace, samples=None, model=None, vars=None, size=None,
     np.random.seed(random_seed)
 
     indices = np.random.randint(0, nchain*len_trace, samples)
-    chain_idx, point_idx = np.divmod(indices, len_trace)
+
     if progressbar:
         indices = tqdm(indices, total=samples)
 
     try:
         ppc = defaultdict(list)
-        for idx in zip(chain_idx, point_idx):
-            param = trace._straces[idx[0]].point(idx[1])
+        for idx in indices:
+            if nchain > 1:
+                chain_idx, point_idx = np.divmod(idx, len_trace)
+                param = trace._straces[chain_idx].point(point_idx)
+            else:
+                param = trace[idx]
+
             for var in vars:
                 ppc[var.name].append(var.distribution.random(point=param,
                                                              size=size))
@@ -1013,8 +1022,9 @@ def sample_ppc_w(traces, samples=None, models=None, weights=None,
 
     Parameters
     ----------
-    traces : list
-        List of traces generated from MCMC sampling. The number of traces should
+    traces : list or list of lists
+        List of traces generated from MCMC sampling, or a list of list
+        containing dicts from find_MAP() or points. The number of traces should
         be equal to the number of weights.
     samples : int
         Number of posterior predictive samples to generate. Defaults to the
@@ -1073,12 +1083,20 @@ def sample_ppc_w(traces, samples=None, models=None, weights=None,
     trace = []
     for i, j in enumerate(n):
         tr = traces[i]
-        len_trace = len(tr) 
-        nchain = tr.nchains 
+        len_trace = len(tr)
+        try:
+            nchain = tr.nchains
+        except AttributeError:
+            nchain = 1
+
         indices = np.random.randint(0, nchain*len_trace, j)
-        chain_idx, point_idx = np.divmod(indices, len_trace)
-        for idx in zip(chain_idx, point_idx): 
-            trace.append(tr._straces[idx[0]].point(idx[1]))
+        if nchain > 1:
+            chain_idx, point_idx = np.divmod(indices, len_trace)
+            for idx in zip(chain_idx, point_idx):
+                trace.append(tr._straces[idx[0]].point(idx[1]))
+        else:
+            for idx in indices:
+                trace.append(tr[idx])
 
     obs = [x for m in models for x in m.observed_RVs]
     variables = np.repeat(obs, n)

pymc3/tests/test_sampling.py

@@ -210,6 +210,8 @@ def test_normal_scalar(self):
             trace = pm.sample()
 
         with model:
+            # test list input
+            ppc0 = pm.sample_ppc([model.test_point], samples=10)
             ppc = pm.sample_ppc(trace, samples=1000, vars=[])
             assert len(ppc) == 0
             ppc = pm.sample_ppc(trace, samples=1000, vars=[a])
@@ -228,6 +230,8 @@ def test_normal_vector(self):
             trace = pm.sample()
 
         with model:
+	    # test list input
+            ppc0 = pm.sample_ppc([model.test_point], samples=10)
             ppc = pm.sample_ppc(trace, samples=10, vars=[])
             assert len(ppc) == 0
             ppc = pm.sample_ppc(trace, samples=10, vars=[a])
@@ -245,6 +249,8 @@ def test_sum_normal(self):
             trace = pm.sample()
 
         with model:
+	    # test list input
+            ppc0 = pm.sample_ppc([model.test_point], samples=10)
             ppc = pm.sample_ppc(trace, samples=1000, vars=[b])
             assert len(ppc) == 1
             assert ppc['b'].shape == (1000,)