shithub: aubio

ref: 007c0a198296077e2adb3bd491fadb0c16538a4c
dir: /python/tests/test_cvec.py/

View raw version
#! /usr/bin/env python

from unittest import main
import numpy as np
from numpy.testing import TestCase, assert_equal
from aubio import cvec, fvec, float_type

wrong_type = 'float32' if float_type == 'float64' else 'float64'

class aubio_cvec_test_case(TestCase):

    def test_vector_created_with_zeroes(self):
        a = cvec(10)
        assert_equal(a.norm.shape[0], 10 // 2 + 1)
        assert_equal(a.phas.shape[0], 10 // 2 + 1)
        assert_equal(a.norm, 0.)
        assert_equal(a.phas, 0.)

    def test_vector_assign_element(self):
        a = cvec()
        a.norm[0] = 1
        assert_equal(a.norm[0], 1)
        a.phas[0] = 1
        assert_equal(a.phas[0], 1)

    def test_vector_assign_element_end(self):
        a = cvec()
        a.norm[-1] = 1
        assert_equal(a.norm[-1], 1)
        assert_equal(a.norm[len(a.norm)-1], 1)
        a.phas[-1] = 1
        assert_equal(a.phas[-1], 1)
        assert_equal(a.phas[len(a.phas)-1], 1)

    def test_assign_cvec_norm_slice(self):
        spec = cvec(1024)
        spec.norm[40:100] = 100
        assert_equal(spec.norm[0:40], 0)
        assert_equal(spec.norm[40:100], 100)
        assert_equal(spec.norm[100:-1], 0)
        assert_equal(spec.phas, 0)

    def test_assign_cvec_phas_slice(self):
        spec = cvec(1024)
        spec.phas[39:-1] = -np.pi
        assert_equal(spec.phas[0:39], 0)
        assert_equal(spec.phas[39:-1], -np.pi)
        assert_equal(spec.norm, 0)

    def test_assign_cvec_with_other_cvec(self):
        """ check dest cvec is still reachable after source was deleted """
        spec = cvec(1024)
        a = np.random.rand(1024//2+1).astype(float_type)
        b = np.random.rand(1024//2+1).astype(float_type)
        spec.norm = a
        spec.phas = b
        new_spec = spec
        del spec
        assert_equal(a, new_spec.norm)
        assert_equal(b, new_spec.phas)
        assert_equal(id(a), id(new_spec.norm))
        assert_equal(id(b), id(new_spec.phas))

    def test_pass_to_numpy(self):
        spec = cvec(1024)
        norm = spec.norm
        phas = spec.phas
        del spec
        new_spec = cvec(1024)
        new_spec.norm = norm
        new_spec.phas = phas
        assert_equal(norm, new_spec.norm)
        assert_equal(phas, new_spec.phas)
        assert_equal(id(norm), id(new_spec.norm))
        assert_equal(id(phas), id(new_spec.phas))
        del norm
        del phas
        assert_equal(new_spec.norm, 0.)
        assert_equal(new_spec.phas, 0.)
        del new_spec

    def test_assign_norm_too_large(self):
        a = cvec(512)
        b = fvec(512//2+1 + 4)
        with self.assertRaises(ValueError):
            a.norm = b

    def test_assign_norm_too_small(self):
        a = cvec(512)
        b = fvec(512//2+1 - 4)
        with self.assertRaises(ValueError):
            a.norm = b

    def test_assign_phas_too_large(self):
        a = cvec(512)
        b = fvec(512//2+1 + 4)
        with self.assertRaises(ValueError):
            a.phas = b

    def test_assign_phas_too_small(self):
        a = cvec(512)
        b = fvec(512//2+1 - 4)
        with self.assertRaises(ValueError):
            a.phas = b

    def test_cvec_repr(self):
        win_s = 512
        c = cvec(win_s)
        expected_repr = "aubio cvec of {:d} elements".format(win_s//2+1)
        self.assertEqual(repr(c), expected_repr)

class aubio_cvec_wrong_norm_input(TestCase):

    def test_wrong_length(self):
        with self.assertRaises(ValueError):
            cvec(-1)

    def test_set_norm_with_scalar(self):
        a = cvec(512)
        with self.assertRaises(ValueError):
            a.norm = 1

    def test_set_norm_with_scalar_array(self):
        a = cvec(512)
        with self.assertRaises(ValueError):
            a.norm = np.ndarray(1, dtype = 'int')

    def test_set_norm_with_int_array(self):
        a = cvec(512)
        with self.assertRaises(ValueError):
            a.norm = np.zeros(512//2+1, dtype = 'int')

    def test_set_norm_with_wrong_float_array(self):
        a = cvec(512)
        with self.assertRaises(ValueError):
            a.norm = np.zeros(512//2+1, dtype = wrong_type)

    def test_set_norm_with_wrong_2d_array(self):
        a = cvec(512)
        with self.assertRaises(ValueError):
            a.norm = np.zeros((512//2+1, 2), dtype = float_type)

if __name__ == '__main__':
    main()