transformations singleton

GPy/core/transformations.py

@@ -9,20 +9,22 @@ lim_val = -np.log(sys.float_info.epsilon)
 
 class Transformation(object):
     domain = None
+    _instance = None
+    def __new__(cls, *args, **kwargs):
+        if not cls._instance:
+            cls._instance = super(Transformation, cls).__new__(
+                                cls, *args, **kwargs)
+        return cls._instance
     def f(self, x):
         raise NotImplementedError
-
     def finv(self, x):
         raise NotImplementedError
-
     def gradfactor(self, f):
         """ df_dx evaluated at self.f(x)=f"""
         raise NotImplementedError
-
     def initialize(self, f):
         """ produce a sensible initial value for f(x)"""
         raise NotImplementedError
-
     def __str__(self):
         raise NotImplementedError
 
@@ -60,6 +62,14 @@ class LogexpClipped(Logexp):
     log_max_bound = np.log(max_bound)
     log_min_bound = np.log(min_bound)
     domain = POSITIVE
+    def __new__(cls, lower=1e-6, *args, **kwargs):
+        if not cls._instance:
+            cls._instance = super(Transformation, cls).__new__(
+                                cls, lower, *args, **kwargs)
+        elif cls._instance.lower == lower:
+            return cls._instance
+        else:
+            return super(Transformation, cls).__new__(cls, lower, *args, **kwargs)
     def __init__(self, lower=1e-6):
         self.lower = lower
     def f(self, x):
@@ -81,6 +91,7 @@ class LogexpClipped(Logexp):
     def __str__(self):
         return '(+ve_c)'
 
+    
 class Exponent(Transformation):
     # TODO: can't allow this to go to zero, need to set a lower bound. Similar with negative Exponent below. See old MATLAB code.
     domain = POSITIVE
@@ -125,10 +136,19 @@ class Square(Transformation):
 
 class Logistic(Transformation):
     domain = BOUNDED
+    def __new__(cls, lower=1e-6, upper=1e-6, *args, **kwargs):
+        if not cls._instance:
+            cls._instance = super(Transformation, cls).__new__(
+                                cls, *args, **kwargs)
+        elif cls._instance.lower == lower and cls._instance.upper == upper:
+            return cls._instance
+        else:
+            return super(Transformation, cls).__new__(cls, *args, **kwargs)
     def __init__(self, lower, upper):
         assert lower < upper
         self.lower, self.upper = float(lower), float(upper)
         self.difference = self.upper - self.lower
+        return self
     def f(self, x):
         return self.lower + self.difference / (1. + np.exp(-x))
     def finv(self, f):
@@ -142,3 +162,5 @@ class Logistic(Transformation):
     def __str__(self):
         return '({},{})'.format(self.lower, self.upper)
 
+ 
+