def invoke_benchmark(batch_size, epochs, num_gpu, mode):
feature_dimension = 10000
train_data = mx.test_utils.rand_ndarray((100000, feature_dimension), 'csr',
0.01)
target_weight = mx.nd.arange(1, feature_dimension + 1).reshape(
(feature_dimension, 1))
train_label = mx.nd.dot(train_data, target_weight)
eval_data = train_data
eval_label = mx.nd.dot(eval_data, target_weight)
train_data = prepare_sliced_sparse_data(train_data, batch_size)
train_label = prepare_sliced_sparse_data(train_label, batch_size)
eval_data = prepare_sliced_sparse_data(eval_data, batch_size)
eval_label = prepare_sliced_sparse_data(eval_label, batch_size)
print("Running Keras benchmark script on sparse data")
keras_sparse_model.run_benchmark(
train_data=sparse.csr_matrix(train_data.asnumpy()),
train_label=train_label.asnumpy(),
eval_data=sparse.csr_matrix(eval_data.asnumpy()),
eval_label=eval_label.asnumpy(),
batch_size=batch_size,
epochs=epochs,
num_gpu=num_gpu,
mode=mode)
def test_prepare_sparse_sliced_data_no_input():
test_train_data = None
batch_size = 3
with pytest.warns(UserWarning): # Warning is thrown when data is None
result = prepare_sliced_sparse_data(test_train_data, batch_size)
assert result is None
def test_prepare_sparse_sliced_data():
test_train_data = _generate_test_data()
batch_size = 2
result = prepare_sliced_sparse_data(test_train_data, batch_size)
assert isinstance(result, sparse.csr.csr_matrix)
assert int(result.shape[0]) % batch_size == 0
def test_prepare_sparse_sliced_data():
row_ind = np.array([0, 1, 1, 3, 4])
col_ind = np.array([0, 2, 4, 3, 4])
data = np.array([1, 2, 3, 4, 5], dtype=float)
test_train_data = sparse.coo_matrix((data, (row_ind, col_ind)))
batch_size = 3
result = prepare_sliced_sparse_data(test_train_data, batch_size)
assert int(result.shape[0]) % batch_size == 0
def test_prepare_sparse_sliced_data_incorrect_dimensions():
row_ind = np.array([0, 1])
col_ind = np.array([0, 2])
data = np.array([1, 2], dtype=float)
test_train_data = sparse.coo_matrix((data, (row_ind, col_ind)))
batch_size = 5
with pytest.warns(UserWarning): # Warning is thrown when data size is smaller than batch size
result = prepare_sliced_sparse_data(test_train_data, batch_size)
assert result.shape[0] == test_train_data.shape[0]
assert result.shape[1] == test_train_data.shape[1]
def test_prepare_sparse_sliced_data_incorrect_dimensions():
test_train_data = _generate_test_data()
batch_size = 5
with pytest.warns(
UserWarning
): # Warning is thrown when data size is smaller than batch size
result = prepare_sliced_sparse_data(test_train_data, batch_size)
assert isinstance(result, sparse.csr.csr_matrix)
assert result.shape[0] == test_train_data.shape[0]
assert result.shape[1] == test_train_data.shape[1]