def test_calculate_radius(self):
map_dim = MapDimension(self.MAP_ROWS, self.MAP_COLS)
total_iter_cnt = torch.tensor(float(1000))
radius = Radius(map_dim, total_iter_cnt)
iter_cnt = 100
r = [radius.radius(i) for i in range(iter_cnt)]
self.assertEqual(len(r), iter_cnt)
def test_provide_neighbourhood_radius(self):
map_dim = MapDimension(10, 15, 3)
total_iter_cnt = torch.tensor(100, dtype=torch.float64)
radius = SomContainer.get_neighbourhood_radius(map_dim, total_iter_cnt)
self.assertIsInstance(radius, Radius)
def setUpClass(cls):
cls.__map_dim = MapDimension(cls.MAP_DIM, cls.MAP_DIM, cls.FEATS_CNT)
cls.__map_indices = MapIndicesBuilder().build(cls.__map_dim)
cls.__bmu = Bmu(EuclideanDistance(), cls.__map_indices)
cls.__total_iter_cnt = torch.tensor(100, dtype=torch.float64)
cls.__learning_rate = LearningRate(0.01, cls.__total_iter_cnt)
cls.__radius = Radius(cls.__map_dim, cls.__total_iter_cnt)
cls.__weights_updater = WeightsUpdater(cls.__map_indices)
def test_get_som(self):
map_dim = MapDimension(10, 10, 3)
board_dir = os.path.join(os.path.dirname(__file__), "board")
train_params = TrainParams(50, 1, 10, 10, board_dir)
som = SomContainer.get_som(map_dim, .001, 100.0, (-1, 1), board_dir,
train_params)
self.assertIsInstance(som, Som)
def test_build_indices(self):
map_dim = MapDimension(5, 10, 2)
map_indices = MapIndicesBuilder()
indices = map_indices.build(map_dim)
self.assertIsInstance(indices, torch.Tensor)
self.assertEqual(indices.shape,
(map_dim.rows, map_dim.cols, map_dim.features))
for row_num in range(map_dim.rows):
for col_num in range(map_dim.cols):
expected_value = np.asarray([row_num, col_num])
np.testing.assert_array_equal(
indices[row_num, col_num].numpy(), expected_value)
def setUpClass(cls):
cls.__logger = cls.__setup_logger()
cls.__corpus_file = cls.__setup_file(cls.FEATS_CNT, 0, cls.ROWS_CNT)
cls.__map_dim = MapDimension(cls.MAP_SIZE, cls.MAP_SIZE, cls.FEATS_CNT)
cls.__map_indices = MapIndicesBuilder().build(cls.__map_dim)
cls.__bmu = Bmu(EuclideanDistance(), cls.__map_indices)
cls.__dataset = CsvDatasetFactory([cls.__corpus_file])
cls.__total_iter_cnt = torch.tensor(cls.ROWS_CNT / cls.BATCH_SIZE,
dtype=torch.float64)
cls.__learning_rate = LearningRate(cls.LEARNING_RATE,
cls.__total_iter_cnt)
cls.__summary_writer = SomSummaryWriter(cls.EVAL_ROOT_DIR)
cls.__radius = Radius(cls.__map_dim, cls.__total_iter_cnt)
cls.__weights_updater = WeightsUpdater(cls.__map_indices)
def setUpClass(cls):
np.set_printoptions(threshold=np.inf)
cls.__corpus_files = cls.__setup_corpus_files()
cls.__logger = cls.__setup_logger()
cls.__map_dim = MapDimension(cls.MAP_SIZE, cls.MAP_SIZE, 3)
cls.__map_indices = MapIndicesBuilder().build(cls.__map_dim)
cls.__bmu = Bmu(EuclideanDistance(), cls.__map_indices)
cls.__dataset = CsvDatasetFactory(cls.__corpus_files, False)
cls.__total_iter_cnt = torch.tensor(cls.SAMPLES_IN_FILE * len(cls.__corpus_files) / cls.BATCH_SIZE,
dtype=torch.float64)
cls.__learning_rate = LearningRate(cls.LEARNING_RATE, cls.__total_iter_cnt)
cls.__radius = Radius(cls.__map_dim, cls.__total_iter_cnt)
cls.__weights_updater = WeightsUpdater(cls.__map_indices)
cls.__weights_plot = WeightsPlot()
cls.__summary_writer = SomSummaryWriter(cls.EVAL_ROOT_DIR)
cls.__weights_sver = WeightsSaver(3)
def test_get_som_model(self):
dist = EuclideanDistance()
map_indices = torch.tensor(np.asarray((10, 10, 2)))
bmu = Bmu(dist, map_indices)
total_iter_cnt = torch.tensor(100, dtype=torch.float64)
learning_rate = LearningRate(.001, total_iter_cnt)
map_dim = MapDimension(10, 10, 3)
radius = Radius(map_dim, total_iter_cnt)
weights_updater = WeightsUpdater(map_indices)
w_range = (-1.0, 1.0)
som_model = SomContainer.get_som_model(bmu, learning_rate, map_dim,
map_indices, radius,
weights_updater, w_range)
self.assertIsInstance(som_model, SomModel)
def setUpClass(cls):
cls.__w_tensor = torch.tensor(cls.W)
cls.__v_tensor = torch.tensor(cls.V)
w_shape = cls.W.shape
cls.__bmu = Bmu(EuclideanDistance(),
MapDimension(w_shape[0], w_shape[1]))