Node output size refactoring (#127)

All outputs are required to be arrays. Make the default output of Prior 1d vector. Do not require any longer that all vectors need to be at least 2d. Strip away unnecessary dimensions if possible to follow numpy convention.

elfi/bo/gpy_regression.py

@@ -136,6 +136,11 @@ def _make_gpy_instance(self, x, y, kernel, noise_var, mean_function):
     def update(self, x, y, optimize=False):
         """Updates the GP model with new data
         """
+
+        # Must cast these as 2d for GPy
+        x = x.reshape((-1, self.input_dim))
+        y = y.reshape((-1, 1))
+
         if self._gp is None:
             self._init_gp(x, y)
         else:

elfi/methods/methods.py

@@ -452,10 +452,10 @@ def _get_batches_total(self):
     def _init_state_samples(self, batch):
         # Initialize the outputs dict based on the received batch
         samples = {}
-        for output in self.outputs:
+        for node in self.outputs:
             shape = (self.objective['n_samples'] + self.batch_size,) \
-                    + batch[output].shape[1:]
-            samples[output] = np.ones(shape) * np.inf
+                    + batch[node].shape[1:]
+            samples[node] = np.ones(shape) * np.inf
         self.state['samples'] = samples
 
     def _merge_batch(self, batch):
@@ -465,8 +465,8 @@ def _merge_batch(self, batch):
         samples = self.state['samples']
 
         # Put the acquired samples to the end
-        for k, v in samples.items():
-            v[self.objective['n_samples']:] = batch[k]
+        for node, v in samples.items():
+            v[self.objective['n_samples']:] = batch[node]
 
         # Sort the smallest to the beginning
         sort_mask = np.argsort(samples[self.discrepancy], axis=0).ravel()

elfi/model/elfi_model.py

@@ -343,15 +343,15 @@ def observed(self):
 
 
 class ScipyLikeRV(StochasticMixin, NodeReference):
-    def __init__(self, distribution="uniform", *params, size=1, **kwargs):
+    def __init__(self, distribution="uniform", *params, size=None, **kwargs):
         """
 
         Parameters
         ----------
         distribution : str or scipy-like distribution object
         params : params of the distribution
-        size : int, tuple or None
-            size of a single random draw. None means a scalar.
+        size : int, tuple or None, optional
+            size of a single random draw. None (default) means a scalar.
 
         """
 

examples/ma2.py

@@ -8,14 +8,15 @@
 """
 
 
-def MA2(t1, t2, n_obs=100, batch_size=1, random_state=None, w=None):
-    if w is None:
-        if random_state is None:
-            # Use the global random_state
-            random_state = np.random
-        w = random_state.randn(batch_size, n_obs+2) # i.i.d. sequence ~ N(0,1)
-
-    w = np.atleast_2d(w)
+def MA2(t1, t2, n_obs=100, batch_size=1, random_state=None):
+    random_state = random_state or np.random
+    # i.i.d. sequence ~ N(0,1)
+    w = random_state.randn(batch_size, n_obs+2)
+
+    # Make inputs 2d arrays for broadcasting with w
+    t1 = np.atleast_2d(t1).reshape((-1, 1))
+    t2 = np.atleast_2d(t2).reshape((-1, 1))
+
     x = w[:, 2:] + t1*w[:, 1:-1] + t2*w[:, :-2]
     return x
 
@@ -24,12 +25,12 @@ def autocov(x, lag=1):
     """Autocovariance assuming a (weak) univariate stationary process with mean 0.
     Realizations are in rows.
     """
-    C = np.mean(x[:, lag:]*x[:, :-lag], axis=1, keepdims=True)
+    C = np.mean(x[:, lag:]*x[:, :-lag], axis=1)
     return C
 
 
 def discrepancy(x, y):
-    d = np.linalg.norm(np.array(x) - np.array(y), ord=2, axis=0)
+    d = np.linalg.norm(np.column_stack(x) - np.column_stack(y), ord=2, axis=1)
     return d
 
 

tests/unit/test_examples.py

@@ -16,8 +16,7 @@ def test_generate():
     res = d.generate(n_gen)
 
     assert res.shape[0] == n_gen
-    assert res.ndim == 2
-    assert res.shape[1] == 1
+    assert res.ndim == 1
 
 
 @pytest.mark.usefixtures('with_all_clients')