Merge branch 'add-test-of-paddle.nn.functional.conv2d_transpose' of h…

…ttps://github.com/SmirnovKol/PaddleTest into add-test-of-paddle.nn.functional.conv2d_transpose

framework/api/nn/test_cosinesimilarity.py

@@ -0,0 +1,266 @@
+#!/bin/env python
+# -*- coding: utf-8 -*-
+# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python
+"""
+test paddle.nn.CosineSimilarity
+"""
+
+from apibase import APIBase
+from apibase import randtool, compare
+import paddle
+import pytest
+import numpy as np
+
+
+class TestCosineSimilarity(APIBase):
+    """
+    test
+    """
+
+    def hook(self):
+        """
+        implement
+        """
+        self.types = [np.float32, np.float64]
+        # self.enable_backward=False
+        # self.debug = True
+        # self.static = True
+        # enable check grad
+
+
+# obj = TestCosineSimilarity(paddle.nn.CosineSimilarity)
+
+
+def cos_sim(x1, x2, axis, eps):
+    """
+    Cosine Similarity using numpy
+    """
+    w12 = np.sum(np.multiply(x1, x2), axis=axis)
+    w1 = np.sum(np.multiply(x1, x1), axis=axis)
+    w2 = np.sum(np.multiply(x2, x2), axis=axis)
+    n12 = np.sqrt(np.clip(w1 * w2, a_min=eps * eps, a_max=None))
+    return w12 / n12
+
+
+@pytest.mark.api_nn_CosineSimilarity_vartype
+def test_cosinesimilarity_base():
+    """
+    base
+    """
+
+    x1 = randtool("float", -10, 10, [1, 5])
+    x2 = randtool("float", -10, 10, [1, 5])
+
+    axis = 1
+    eps = 1e-8
+
+    res = cos_sim(x1, x2, axis, eps)
+
+    x1 = paddle.to_tensor(x1)
+    x2 = paddle.to_tensor(x2)
+
+    cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+    paddle_res = cos_sim_func(x1, x2).numpy()
+
+    compare(paddle_res, res)
+
+
+@pytest.mark.api_nn_CosineSimilarity_parameters
+def test_cosinesimilarity():
+    """
+    input shape
+    test on 3_D input
+    """
+    x1 = randtool("float", -10, 10, [3, 8, 2])
+    x2 = randtool("float", -10, 10, [3, 8, 2])
+
+    axis = 1
+    eps = 1e-8
+
+    res = cos_sim(x1, x2, axis, eps)
+
+    x1 = paddle.to_tensor(x1)
+    x2 = paddle.to_tensor(x2)
+
+    cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+    paddle_res = cos_sim_func(x1, x2).numpy()
+
+    compare(paddle_res, res)
+
+
+@pytest.mark.api_nn_CosineSimilarity_parameters
+def test_cosinesimilarity2():
+    """
+    axis (int): Dimension of vectors to compute cosine similarity. Default is 1.
+    test on -1, 0, 1
+    """
+    x1 = randtool("float", -10, 10, [1, 5, 2])
+    x2 = randtool("float", -10, 10, [1, 5, 2])
+
+    axis_list = [-2, -1, 0, 1, 2]
+    eps = 1e-8
+
+    for axis in axis_list:
+        res = cos_sim(x1, x2, axis, eps)
+
+        x1_tensor = paddle.to_tensor(x1)
+        x2_tensor = paddle.to_tensor(x2)
+
+        cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+        paddle_res = cos_sim_func(x1_tensor, x2_tensor).numpy()
+
+        compare(paddle_res, res)
+
+
+@pytest.mark.api_nn_CosineSimilarity_parameters
+def test_cosinesimilarity3():
+    """
+    eps(float): Small value to avoid division by zero. Default is 1e-8.
+    test on 1e-6, 1e-6, 1e-7, 1e-8, 1e-9
+    """
+    x1 = randtool("float", -10, 10, [1, 5])
+    x2 = randtool("float", -10, 10, [1, 5])
+
+    axis = 1
+    eps_list = [1e-6, 1e-6, 1e-7, 1e-8, 1e-9]
+
+    for eps in eps_list:
+        res = cos_sim(x1, x2, axis, eps)
+
+        x1_tensor = paddle.to_tensor(x1)
+        x2_tensor = paddle.to_tensor(x2)
+
+        cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+        paddle_res = cos_sim_func(x1_tensor, x2_tensor).numpy()
+
+        compare(paddle_res, res)
+
+
+@pytest.mark.api_nn_CosineSimilarity_parameters
+def test_cosinesimilarity4():
+    """
+    input value
+    test on different shape
+    """
+    x1 = randtool("float", -100, 100, [5, 2, 3, 4])
+    x2 = randtool("float", -100, 100, [2, 3, 4])
+
+    axis = -1
+    eps = 1e-8
+
+    res = cos_sim(x1, x2, axis, eps)
+
+    x1_tensor = paddle.to_tensor(x1)
+    x2_tensor = paddle.to_tensor(x2)
+
+    cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+    paddle_res = cos_sim_func(x1_tensor, x2_tensor).numpy()
+
+    compare(paddle_res, res)
+
+
+@pytest.mark.api_nn_CosineSimilarity_parameters
+def test_cosinesimilarity5():
+    """
+    input value
+    test on different shape and axis set at 0
+    """
+    x1 = randtool("float", -100, 100, [5, 1, 3])
+    x2 = randtool("float", -100, 100, [1, 3])
+
+    axis = 0
+    eps = 1e-8
+
+    res = cos_sim(x1, x2, axis, eps)
+
+    x1_tensor = paddle.to_tensor(x1)
+    x2_tensor = paddle.to_tensor(x2)
+
+    cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+    paddle_res = cos_sim_func(x1_tensor, x2_tensor).numpy()
+
+    compare(paddle_res, res)
+    # obj.run(res=res,x1=x1,x2=x2)
+
+
+@pytest.mark.api_nn_CosineSimilarity_parameters
+def test_cosinesimilarity6():
+    """
+    error axis
+    axis out of range
+    """
+    x1 = randtool("float", -100, 100, [5, 1, 3])
+    x2 = randtool("float", -100, 100, [1, 3])
+
+    axis = 2
+    eps = 1e-8
+
+    x1_tensor = paddle.to_tensor(x1)
+    x2_tensor = paddle.to_tensor(x2)
+
+    cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+
+    # cos_sim_func(x1_tensor, x2_tensor)
+    try:
+        cos_sim_func(x1_tensor, x2_tensor)
+    except Exception as e:
+        # print(e)
+        if "[operator < reduce_sum > error]" in e.args[0]:
+            pass
+        else:
+            raise Exception
+
+
+@pytest.mark.api_nn_CosineSimilarity_parameters
+def test_cosinesimilarity7():
+    """
+    error type
+    type is not float
+    """
+    x1 = randtool("int", -100, 100, [5, 1, 3])
+    x2 = randtool("float", -100, 100, [5, 1, 3])
+
+    axis = 1
+    eps = 1e-8
+
+    x1_tensor = paddle.to_tensor(x1)
+    x2_tensor = paddle.to_tensor(x2)
+
+    cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+
+    # cos_sim_func(x1_tensor, x2_tensor)
+    try:
+        cos_sim_func(x1_tensor, x2_tensor)
+    except Exception as e:
+        # print(e)
+        if "[operator < elementwise_mul > error]" in e.args[0]:
+            pass
+        else:
+            raise Exception
+
+
+def test_cosinesimilarity8():
+    """
+    error shape
+    Broadcast dimension mismatch
+    """
+    x1 = randtool("float", -100, 100, [5, 1, 3])
+    x2 = randtool("float", -100, 100, [2, 3])
+
+    axis = 1
+    eps = 1e-8
+
+    x1_tensor = paddle.to_tensor(x1)
+    x2_tensor = paddle.to_tensor(x2)
+
+    cos_sim_func = paddle.nn.CosineSimilarity(axis=axis, eps=eps)
+
+    # cos_sim_func(x1_tensor, x2_tensor)
+    try:
+        cos_sim_func(x1_tensor, x2_tensor)
+    except Exception as e:
+        # print(e)
+        if "[operator < elementwise_mul > error]" in e.args[0]:
+            pass
+        else:
+            raise Exception