!9050 Mix bugs in Categorical distribution

From: @shallydeng
Reviewed-by: @zichun_ye,@sunnybeike
Signed-off-by: @sunnybeike

mindspore/nn/probability/distribution/_utils/utils.py

@@ -285,11 +285,11 @@ class CheckTuple(PrimitiveWithInfer):
         return out
 
     def __call__(self, x, name):
-        if context.get_context("mode") == 0:
-            return x["value"]
-        # Pynative mode
+        # The op is not used in a cell
         if isinstance(x, tuple):
             return x
+        if context.get_context("mode") == 0:
+            return x["value"]
         raise TypeError(f"For {name}, input type should be a tuple.")
 
 

mindspore/nn/probability/distribution/categorical.py

@@ -108,7 +108,7 @@ class Categorical(Distribution):
                  name="Categorical"):
         param = dict(locals())
         param['param_dict'] = {'probs': probs}
-        valid_dtype = mstype.int_type
+        valid_dtype = mstype.int_type + mstype.float_type
         Validator.check_type_name("dtype", dtype, valid_dtype, type(self).__name__)
         super(Categorical, self).__init__(seed, dtype, name, param)
 
@@ -134,8 +134,8 @@ class Categorical(Distribution):
         self.exp = exp_generic
         self.expand_dim = P.ExpandDims()
         self.fill = P.Fill()
-        self.floor = P.Floor()
         self.gather = P.GatherNd()
+        self.issubclass = P.IsSubClass()
         self.less = P.Less()
         self.log = log_generic
         self.log_softmax = P.LogSoftmax()
@@ -153,6 +153,7 @@ class Categorical(Distribution):
         self.transpose = P.Transpose()
 
         self.index_type = mstype.int32
+        self.nan = np.nan
 
 
     def extend_repr(self):
@@ -255,7 +256,10 @@ class Categorical(Distribution):
             probs (Tensor): Event probabilities. Default: self.probs.
         """
         value = self._check_value(value, 'value')
-        value = self.cast(value, self.parameter_type)
+        if self.issubclass(self.dtype, mstype.float_):
+            value = self.cast(value, self.index_type)
+        else:
+            value = self.cast(value, self.dtype)
         probs = self._check_param_type(probs)
         logits = self.log(probs)
 
@@ -294,8 +298,8 @@ class Categorical(Distribution):
         # index into logit_pmf, fill in out_of_bound places with -inf
         # reshape into label shape N
         logits_pmf = self.gather(self.reshape(logits, (-1, num_classes)), index)
-        neg_inf = self.fill(self.dtypeop(logits_pmf), self.shape(logits_pmf), -np.inf)
-        logits_pmf = self.select(out_of_bound, neg_inf, logits_pmf)
+        nan = self.fill(self.dtypeop(logits_pmf), self.shape(logits_pmf), self.nan)
+        logits_pmf = self.select(out_of_bound, nan, logits_pmf)
         ans = self.reshape(logits_pmf, label_shape)
         if drop_dim:
             return self.squeeze(ans)
@@ -310,8 +314,10 @@ class Categorical(Distribution):
             probs (Tensor): Event probabilities. Default: self.probs.
         """
         value = self._check_value(value, 'value')
-        value = self.cast(value, self.parameter_type)
-        value = self.floor(value)
+        if self.issubclass(self.dtype, mstype.float_):
+            value = self.cast(value, self.index_type)
+        else:
+            value = self.cast(value, self.dtype)
         probs = self._check_param_type(probs)
 
         # handle the case when value is of shape () and probs is a scalar batch

tests/st/probability/distribution/test_categorical.py

@@ -219,7 +219,7 @@ def test_log_survival():
     Test log survival funciton.
     """
     expect_logsurvival = np.log([1., 0.3, 0.3, 0.3, 0.3])
-    x_ = Tensor(np.array([-0.1, 0, 0, 0.5, 0.5]).astype(np.float32), dtype=dtype.float32)
+    x_ = Tensor(np.array([-2, 0, 0, 0.5, 0.5]).astype(np.float32), dtype=dtype.float32)
     log_sf = LogSF()
     output = log_sf(x_)
     tol = 1e-6