make test_hsigmoid_op.py right

python/paddle/fluid/tests/unittests/test_hsigmoid_op.py

-# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,8 +14,8 @@
 
 import unittest
 import numpy as np
-from op_test import OpTest
 import math
+from op_test import OpTest
 
 
 def find_latest_set(num):
@@ -37,40 +37,36 @@ class CodeTable(object):
 
 
 def hsigmoid(x, w, ids, bias, num_classes):
-    # code length = 
-    # initialize pre out with dims={batch_size, code_length}
+    global pre_output
     batch_size = x.shape[0]
     code_length = find_latest_set(num_classes - 1)
     code_table = [0 for _ in range(code_length)]
     pre_output = np.zeros((batch_size, code_length))
     pre_sum = np.zeros((batch_size, 1))
     out = np.zeros((batch_size, 1)).astype("float32")
-    # pre_out += code(bias)
-    for i in xrange(batch_size):
+    for i in range(batch_size):
         code_table = CodeTable(num_classes, ids[i])
         length = code_table.get_length()
-        for j in xrange(length):
+        for j in range(length):
             idx = code_table.cal_index(j)
             pre_output[i][j] += bias[0][idx]
-    # pre_out += code(w) * x
-    for i in xrange(batch_size):
-        for j in xrange(batch_size):
-            code_table = CodeTable(num_classes, ids[j])
-            length = code_table.get_length()
-            for k in xrange(length):
-                idx = code_table.cal_index(k)
-                sum = 0.0
-                for l in xrange(x.shape[1]):
-                    sum += w[i][idx][l] * x[j][l]
-                pre_output[j][k] += sum
+    for j in range(batch_size):
+        code_table = CodeTable(num_classes, ids[j])
+        length = code_table.get_length()
+        for k in range(length):
+            idx = code_table.cal_index(k)
+            sum = 0.0
+            for l in range(x.shape[1]):
+                sum += w[idx][l] * x[j][l]
+            pre_output[j][k] += sum
     # clip[-40.0, 40.0]
     np.clip(pre_output, -40.0, 40.0)
     # out(i, 0) = \sum_j  bit(i, j) * preout(i, j)
-    for i in xrange(batch_size):
+    for i in range(batch_size):
         code_table = CodeTable(num_classes, ids[i])
         length = code_table.get_length()
         sum = 0.0
-        for j in xrange(length):
+        for j in range(length):
             if code_table.cal_bit(j):
                 sum += pre_output[i][j]
         out[i] = -1.0 * sum
@@ -85,24 +81,23 @@ def hsigmoid(x, w, ids, bias, num_classes):
 class TestHSigmoidOp(OpTest):
     def setUp(self):
         self.op_type = "hierarchical_sigmoid"
-        num_classes = 6
-        embded_size = 10
-        batch_size = 5
+        num_classes = 4
+        embded_size = 1
+        batch_size = 1
         x = np.random.random((batch_size, embded_size)).astype("float32")
-        w = np.random.random(
-            (batch_size, num_classes - 1, embded_size)).astype("float32")
+        w = np.random.random((num_classes - 1, embded_size)).astype("float32")
         ids = np.random.randint(0, num_classes, batch_size)
         bias = np.random.random((1, num_classes - 1)).astype("float32")
-        self.inputs = {'X': x, 'W': w, 'Ids': ids, 'Bias': bias}
         self.attrs = {'num_classes': num_classes}
+        self.inputs = {'X': x, 'W': w, 'Ids': ids, 'Bias': bias}
         out = hsigmoid(x, w, ids, bias, num_classes)
-        self.outputs = {'Out': out}
+        self.outputs = {'PreOut': pre_output, 'Out': out}
 
     def test_check_output(self):
         self.check_output()
 
     def test_check_grad(self):
-        self.check_grad(['X', 'W', 'Bias'], 'Out', no_grad_set=set('Ids'))
+        self.check_grad(['Bias', 'X', 'W'], 'Out', no_grad_set=set('Ids'))
 
 
 if __name__ == '__main__':