REF: norms _set_tuning_param default to inplace=False

statsmodels/robust/norms.py

@@ -202,12 +202,16 @@ class HuberT(RobustNorm):
     def __init__(self, t=1.345):
         self.t = t
 
-    def _set_tuning_param(self, c):
+    def _set_tuning_param(self, c, inplace=False):
         """Set and change the tuning parameter of the Norm.
 
         Warning: this needs to wipe cached attributes that depend on the param.
         """
-        self.t = c
+        if inplace:
+            self.t = c
+            return self
+        else:
+            return self.__class__(t=c)
 
     def max_rho(self):
         return np.inf
@@ -325,6 +329,18 @@ class RamsayE(RobustNorm):
     def __init__(self, a=.3):
         self.a = a
 
+    def _set_tuning_param(self, c, inplace=False):
+        """Set and change the tuning parameter of the Norm.
+
+        Warning: this needs to wipe cached attributes that depend on the param.
+        """
+        # todo : change default to inplace=False, when tools are fixed
+        if inplace:
+            self.a = c
+            return self
+        else:
+            return self.__class__(a=c)
+
     def max_rho(self):
         return np.inf
 
@@ -422,12 +438,16 @@ class AndrewWave(RobustNorm):
     def __init__(self, a=1.339):
         self.a = a
 
-    def _set_tuning_param(self, a):
+    def _set_tuning_param(self, c, inplace=False):
         """Set and change the tuning parameter of the Norm.
 
         Warning: this needs to wipe cached attributes that depend on the param.
         """
-        self.a = a
+        if inplace:
+            self.a = c
+            return self
+        else:
+            return self.__class__(a=c)
 
     def max_rho(self):
         return 2 * self.a**2
@@ -556,12 +576,16 @@ class TrimmedMean(RobustNorm):
     def __init__(self, c=2.):
         self.c = c
 
-    def _set_tuning_param(self, c):
+    def _set_tuning_param(self, c, inplace=False):
         """Set and change the tuning parameter of the Norm.
 
         Warning: this needs to wipe cached attributes that depend on the param.
         """
-        self.c = c
+        if inplace:
+            self.c = c
+            return self
+        else:
+            return self.__class__(c=c)
 
     def max_rho(self):
         return self.rho(self.c)
@@ -677,14 +701,20 @@ def __init__(self, a=2., b=4., c=8.):
         self.b = b
         self.c = c
 
-    def _set_tuning_param(self, c):
+    def _set_tuning_param(self, c, inplace=False):
         """Set and change the tuning parameter of the Norm.
 
         Warning: this needs to wipe cached attributes that depend on the param.
         """
-        self.c = c
-        self.a = c / 4
-        self.b = c / 2
+        a = c / 4
+        b = c / 2
+        if inplace:
+            self.c = c
+            self.a = a
+            self.b = b
+            return self
+        else:
+            return self.__class__(a=a, b=b, c=c)
 
     def max_rho(self):
         return self.rho(self.c)
@@ -897,7 +927,7 @@ def get_tuning(cls, bp=None, eff=None):
         elif eff is not None:
             return rtools.tukeybiweight_eff[eff]
 
-    def _set_tuning_param(self, c, inplace=True):
+    def _set_tuning_param(self, c, inplace=False):
         """Set and change the tuning parameter of the Norm.
 
         Warning: this needs to wipe cached attributes that depend on the param.
@@ -1023,12 +1053,16 @@ def __init__(self, c=3.61752, k=4):
         self.c = c
         self.k = k
 
-    def _set_tuning_param(self, c):
+    def _set_tuning_param(self, c, inplace=False):
         """Set and change the tuning parameter of the Norm.
 
         Warning: this needs to wipe cached attributes that depend on the param.
         """
-        self.c = c
+        if inplace:
+            self.c = c
+            return self
+        else:
+            return self.__class__(c=c, k=self.k)
 
     def max_rho(self):
         return self.rho(self.c)
@@ -1161,12 +1195,16 @@ def __init__(self, c=2.3849, df=4):
         self.c = c
         self.df = df
 
-    def _set_tuning_param(self, c):
+    def _set_tuning_param(self, c, inplace=False):
         """Set and change the tuning parameter of the Norm.
 
         Warning: this needs to wipe cached attributes that depend on the param.
         """
-        self.c = c
+        if inplace:
+            self.c = c
+            return self
+        else:
+            return self.__class__(c=c, df=self.df)
 
     def max_rho(self):
         return np.inf

statsmodels/robust/tests/test_tools.py

@@ -57,14 +57,14 @@ def test_eff(case):
 
     res_eff = []
     for c in res2:
-        norm._set_tuning_param(c)
+        norm._set_tuning_param(c, inplace=True)
         res_eff.append(1 / var_func(norm))
 
     assert_allclose(res_eff, effs, atol=0.0005)
 
     for c in res2:
         # bp = stats.norm.expect(lambda x : norm.rho(x)) / norm.rho(norm.c)
-        norm._set_tuning_param(c)
+        norm._set_tuning_param(c, inplace=True)
         eff = 1 / _var_normal(norm)
         tune = _get_tuning_param(norm, eff)
         assert_allclose(tune, c, rtol=1e-6, atol=5e-4)

statsmodels/robust/tools.py

@@ -150,11 +150,11 @@ def _get_tuning_param(norm, eff, kwd="c", kwargs=None, use_jump=False,
         def func(c):
             # kwds.update({kwd: c})
             # return _var_normal(norm(**kwds)) - 1 / eff
-            norm._set_tuning_param(c)
+            norm._set_tuning_param(c, inplace=True)
             return _var_normal(norm) - 1 / eff
     else:
         def func(c):
-            norm._set_tuning_param(c)
+            norm._set_tuning_param(c, inplace=True)
             return _var_normal_jump(norm(**kwds) - 1 / eff)
 
     res = optimize.brentq(func, *bracket)
@@ -215,14 +215,14 @@ def tuning_s_estimator_mean(norm, breakdown=None):
 
     def func(c):
         norm_ = norm
-        norm_._set_tuning_param(c)
+        norm_._set_tuning_param(c, inplace=True)
         bp = stats.norm.expect(lambda x : norm_.rho(x)) / norm_.max_rho()
         return bp
 
     res = []
     for bp in bps:
         c_bp = optimize.brentq(lambda c0: func(c0) - bp, 0.1, 10)
-        norm._set_tuning_param(c_bp)  # inplace modification
+        norm._set_tuning_param(c_bp, inplace=True)  # inplace modification
         eff = 1 / _var_normal(norm)
         b = stats.norm.expect(lambda x : norm.rho(x))
         res.append([bp, eff, c_bp, b])