Probabilty inference for arc transformations

Co-authored-by: Luke LB <ll17354@bristol.ac.uk>

pymc/logprob/transforms.py

@@ -53,6 +53,9 @@
 from pytensor.scalar import (
     Abs,
     Add,
+    ArcCosh,
+    ArcSinh,
+    ArcTanh,
     Cosh,
     Erf,
     Erfc,
@@ -71,6 +74,9 @@
 from pytensor.tensor.math import (
     abs,
     add,
+    arccosh,
+    arcsinh,
+    arctanh,
     cosh,
     erf,
     erfc,
@@ -369,7 +375,23 @@ def apply(self, fgraph: FunctionGraph):
 class MeasurableTransform(MeasurableElemwise):
     """A placeholder used to specify a log-likelihood for a transformed measurable variable"""
 
-    valid_scalar_types = (Exp, Log, Add, Mul, Pow, Abs, Sinh, Cosh, Tanh, Erf, Erfc, Erfcx)
+    valid_scalar_types = (
+        Exp,
+        Log,
+        Add,
+        Mul,
+        Pow,
+        Abs,
+        Sinh,
+        Cosh,
+        Tanh,
+        ArcSinh,
+        ArcCosh,
+        ArcTanh,
+        Erf,
+        Erfc,
+        Erfcx,
+    )
 
     # Cannot use `transform` as name because it would clash with the property added by
     # the `TransformValuesRewrite`
@@ -501,7 +523,9 @@ def measurable_sub_to_neg(fgraph, node):
     return [pt.add(minuend, pt.neg(subtrahend))]
 
 
-@node_rewriter([exp, log, add, mul, pow, abs, sinh, cosh, tanh, erf, erfc, erfcx])
+@node_rewriter(
+    [exp, log, add, mul, pow, abs, sinh, cosh, tanh, arcsinh, arccosh, arctanh, erf, erfc, erfcx]
+)
 def find_measurable_transforms(fgraph: FunctionGraph, node: Node) -> Optional[List[Node]]:
     """Find measurable transformations from Elemwise operators."""
 
@@ -544,6 +568,9 @@ def find_measurable_transforms(fgraph: FunctionGraph, node: Node) -> Optional[Li
         Sinh: SinhTransform(),
         Cosh: CoshTransform(),
         Tanh: TanhTransform(),
+        ArcSinh: ArcsinhTransform(),
+        ArcCosh: ArccoshTransform(),
+        ArcTanh: ArctanhTransform(),
         Erf: ErfTransform(),
         Erfc: ErfcTransform(),
         Erfcx: ErfcxTransform(),
@@ -660,6 +687,39 @@ def backward(self, value, *inputs):
         return pt.arctanh(value)
 
 
+class ArcsinhTransform(RVTransform):
+    name = "arcsinh"
+    ndim_supp = 0
+
+    def forward(self, value, *inputs):
+        return pt.arcsinh(value)
+
+    def backward(self, value, *inputs):
+        return pt.sinh(value)
+
+
+class ArccoshTransform(RVTransform):
+    name = "arccosh"
+    ndim_supp = 0
+
+    def forward(self, value, *inputs):
+        return pt.arccosh(value)
+
+    def backward(self, value, *inputs):
+        return pt.cosh(value)
+
+
+class ArctanhTransform(RVTransform):
+    name = "arctanh"
+    ndim_supp = 0
+
+    def forward(self, value, *inputs):
+        return pt.arctanh(value)
+
+    def backward(self, value, *inputs):
+        return pt.tanh(value)
+
+
 class ErfTransform(RVTransform):
     name = "erf"
     ndim_supp = 0

tests/logprob/test_transforms.py

@@ -51,6 +51,9 @@
 from pymc.logprob.abstract import MeasurableVariable, _logprob
 from pymc.logprob.basic import conditional_logp, logp
 from pymc.logprob.transforms import (
+    ArccoshTransform,
+    ArcsinhTransform,
+    ArctanhTransform,
     ChainedTransform,
     CoshTransform,
     ErfcTransform,
@@ -1028,6 +1031,9 @@ def test_multivariate_transform(shift, scale):
         (pt.sinh, SinhTransform()),
         (pt.cosh, CoshTransform()),
         (pt.tanh, TanhTransform()),
+        (pt.arcsinh, ArcsinhTransform()),
+        (pt.arccosh, ArccoshTransform()),
+        (pt.arctanh, ArctanhTransform()),
     ],
 )
 def test_erf_logp(pt_transform, transform):
@@ -1060,6 +1066,9 @@ def test_erf_logp(pt_transform, transform):
         SinhTransform(),
         CoshTransform(),
         TanhTransform(),
+        ArcsinhTransform(),
+        ArccoshTransform(),
+        ArctanhTransform(),
     ],
 )
 def test_check_jac_det(transform):