Chain transformation

RELEASE-NOTES.md

@@ -8,6 +8,7 @@
 
 - Add `check_test_point` method to `pm.Model`
 - Add `Ordered` Transformation and `OrderedLogistic` distribution
+- Add `Chain` transformation
 - Better warning message for `Mass matrix contains zeros on the diagonal. Some derivatives might always be zero`
 
 ### Fixes

pymc3/distributions/transforms.py

@@ -381,7 +381,7 @@ def jacobian_det(self, x):
 
 
 class CholeskyCovPacked(Transform):
-    name = "cholesky_cov_packed"
+    name = "cholesky-cov-packed"
 
     def __init__(self, n):
         self.diag_idxs = np.arange(1, n + 1).cumsum() - 1
@@ -398,3 +398,45 @@ def forward_val(self, y, point=None):
 
     def jacobian_det(self, y):
         return tt.sum(y[self.diag_idxs])
+
+
+class Chain(Transform):
+    def __init__(self, transform_list):
+        self.transform_list = transform_list
+        self.name = '+'.join([transf.name for transf in self.transform_list])
+
+    def forward(self, x):
+        y = x
+        for transf in self.transform_list:
+            y = transf.forward(y)
+        return y
+
+    def forward_val(self, x, point=None):
+        y = x
+        for transf in self.transform_list:
+            y = transf.forward_val(y)
+        return y
+
+    def backward(self, y):
+        x = y
+        for transf in reversed(self.transform_list):
+            x = transf.backward(x)
+        return x
+
+    def jacobian_det(self, y):
+        y = tt.as_tensor_variable(y)
+        det_list = []
+        ndim0 = y.ndim
+        for transf in reversed(self.transform_list):
+            det_ = transf.jacobian_det(y)
+            det_list.append(det_)
+            y = transf.backward(y)
+            ndim0 = min(ndim0, det_.ndim)
+        # match the shape of the smallest jacobian_det
+        det = 0.
+        for det_ in det_list:
+            if det_.ndim > ndim0:
+                det += det_.sum(axis=-1)
+            else:
+                det += det_
+        return det

pymc3/tests/test_distributions.py

@@ -222,6 +222,14 @@ def SortedVector(n):
     return Domain(vals, edges=(None, None))
 
 
+def UnitSortedVector(n):
+    vals = []
+    np.random.seed(42)
+    for _ in range(10):
+        vals.append(np.sort(np.random.rand(n)))
+    return Domain(vals, edges=(None, None))
+
+
 def RealMatrix(n, m):
     vals = []
     np.random.seed(42)