From 09328382e926b92a42ba5b3df6f169f825975d88 Mon Sep 17 00:00:00 2001 From: reminisce Date: Wed, 5 Jul 2017 11:05:41 -0700 Subject: [PATCH] Revert "Change storage_type to stype" This reverts commit 90db7d18b624f3ee4ffd37bf5680205e77ca2763. --- python/mxnet/_ctypes/ndarray.py | 4 +-- python/mxnet/module/module.py | 2 +- python/mxnet/ndarray.py | 4 +-- python/mxnet/ndarray_utils.py | 20 ++++++------- python/mxnet/optimizer.py | 2 +- python/mxnet/sparse_ndarray.py | 8 +++--- python/mxnet/symbol.py | 18 ++++++------ python/mxnet/test_utils.py | 18 ++++++------ src/operator/tensor/cast_storage-inl.h | 8 +++--- tests/python/unittest/test_kvstore.py | 4 +-- tests/python/unittest/test_module.py | 12 ++++---- .../python/unittest/test_multi_device_exec.py | 4 +-- tests/python/unittest/test_sparse_ndarray.py | 28 +++++++++---------- tests/python/unittest/test_sparse_operator.py | 28 +++++++++---------- 14 files changed, 79 insertions(+), 81 deletions(-) diff --git a/python/mxnet/_ctypes/ndarray.py b/python/mxnet/_ctypes/ndarray.py index 1572d8f8898d..494db9356271 100644 --- a/python/mxnet/_ctypes/ndarray.py +++ b/python/mxnet/_ctypes/ndarray.py @@ -93,8 +93,8 @@ def _imperative_invoke(handle, ndargs, keys, vals, out): return _ndarray_cls(ctypes.cast(output_vars[0], NDArrayHandle), stype=_STORAGE_TYPE_ID_TO_STR[out_stypes[0]]) else: - return [_ndarray_cls(ctypes.cast(output_vars[i], NDArrayHandle), - stype=_STORAGE_TYPE_ID_TO_STR[out_stypes[i]]) + return [_ndarray_cls(ctypes.cast(output_vars[i], NDArrayHandle, + stype=_STORAGE_TYPE_ID_TO_STR[out_stypes[i]])) for i in range(num_output.value)] diff --git a/python/mxnet/module/module.py b/python/mxnet/module/module.py index bd66c8eef34e..680b937d3fd3 100644 --- a/python/mxnet/module/module.py +++ b/python/mxnet/module/module.py @@ -399,7 +399,7 @@ def bind(self, data_shapes, label_shapes=None, for_training=True, else: assert self._arg_params is None and self._aux_params is None param_arrays = [ - mx.nd.zeros(shape=x[0].shape, dtype=x[0].dtype, stype=x[0].stype) + mx.nd.zeros(shape=x[0].shape, dtype=x[0].dtype, storage_type=x[0].stype) for x in self._exec_group.param_arrays ] self._arg_params = {name:arr for name, arr in zip(self._param_names, param_arrays)} diff --git a/python/mxnet/ndarray.py b/python/mxnet/ndarray.py index 53082046fad7..948d8dd587e7 100644 --- a/python/mxnet/ndarray.py +++ b/python/mxnet/ndarray.py @@ -973,11 +973,11 @@ def backward(self, out_grad=None, retain_graph=False): def _to_csr(self): # pylint: disable=undefined-variable - return cast_storage(self, stype='csr') + return cast_storage(self, storage_type='csr') def _to_rsp(self): # pylint: disable=undefined-variable - return cast_storage(self, stype='row_sparse') + return cast_storage(self, storage_type='row_sparse') def onehot_encode(indices, out): """One-hot encoding indices into matrix out. diff --git a/python/mxnet/ndarray_utils.py b/python/mxnet/ndarray_utils.py index a6a71bf25f6f..5f8fa6c7bfb7 100644 --- a/python/mxnet/ndarray_utils.py +++ b/python/mxnet/ndarray_utils.py @@ -48,14 +48,14 @@ def _zeros_ndarray(shape, ctx=None, dtype=None, **kwargs): # pylint: enable= no-member, protected-access -def _zeros_sparse_ndarray(stype, shape, ctx=None, dtype=None, aux_types=None, **kwargs): +def _zeros_sparse_ndarray(storage_type, shape, ctx=None, dtype=None, aux_types=None, **kwargs): """Return a new array of given shape and type, filled with zeros. Parameters ---------- shape : int or tuple of int The shape of the empty array - stype: string + storage_type: string The storage type of the empty array, such as 'row_sparse', 'csr', etc ctx : Context, optional An optional device context (default is the current default context) @@ -76,26 +76,26 @@ def _zeros_sparse_ndarray(stype, shape, ctx=None, dtype=None, aux_types=None, ** >>> mx.sparse_nd.zeros('row_sparse', (1,2), mx.gpu(0), 'float16').asnumpy() array([[ 0., 0.]], dtype=float16) """ - if stype == 'default': + if storage_type == 'default': return _zeros_ndarray(shape, ctx=ctx, dtype=dtype, **kwargs) if ctx is None: ctx = Context.default_ctx dtype = mx_real_t if dtype is None else dtype if aux_types is None: - if stype == 'row_sparse' or stype == 'csr': - aux_types = _STORAGE_AUX_TYPES[stype] + if storage_type == 'row_sparse' or storage_type == 'csr': + aux_types = _STORAGE_AUX_TYPES[storage_type] else: raise Exception("unknown storage type") - assert(len(aux_types) == len(_STORAGE_AUX_TYPES[stype])) - out = _ndarray_cls(_new_alloc_handle(stype, shape, ctx, True, dtype, aux_types)) + assert(len(aux_types) == len(_STORAGE_AUX_TYPES[storage_type])) + out = _ndarray_cls(_new_alloc_handle(storage_type, shape, ctx, True, dtype, aux_types)) return _internal._zeros(shape=shape, ctx=ctx, dtype=dtype, out=out, **kwargs) -def zeros(shape, ctx=None, dtype=None, stype=None, aux_types=None, **kwargs): - if stype is None: +def zeros(shape, ctx=None, dtype=None, storage_type=None, aux_types=None, **kwargs): + if storage_type is None: return _zeros_ndarray(shape, ctx, dtype, **kwargs) else: - return _zeros_sparse_ndarray(stype, shape, ctx, dtype, aux_types, **kwargs) + return _zeros_sparse_ndarray(storage_type, shape, ctx, dtype, aux_types, **kwargs) def load(fname): diff --git a/python/mxnet/optimizer.py b/python/mxnet/optimizer.py index 5ee50e755afc..7e46c30c7c79 100644 --- a/python/mxnet/optimizer.py +++ b/python/mxnet/optimizer.py @@ -335,7 +335,7 @@ def create_state(self, index, weight): return None else: return mx.nd.zeros(shape=weight.shape, ctx=weight.context, - dtype=weight.dtype, stype=weight.stype) + dtype=weight.dtype, storage_type=weight.stype) def update(self, index, weight, grad, state): assert(isinstance(weight, NDArray)) diff --git a/python/mxnet/sparse_ndarray.py b/python/mxnet/sparse_ndarray.py index f2daa5c8fb0a..a438b4d6ec7d 100644 --- a/python/mxnet/sparse_ndarray.py +++ b/python/mxnet/sparse_ndarray.py @@ -48,7 +48,7 @@ } -def _new_alloc_handle(stype, shape, ctx, delay_alloc, dtype, aux_types, aux_shapes=None): +def _new_alloc_handle(storage_type, shape, ctx, delay_alloc, dtype, aux_types, aux_shapes=None): """Return a new handle with specified storage type, shape, dtype and context. Empty handle is only used to hold results @@ -65,7 +65,7 @@ def _new_alloc_handle(stype, shape, ctx, delay_alloc, dtype, aux_types, aux_shap aux_shapes = sum(aux_shapes, ()) num_aux = mx_uint(len(aux_types)) check_call(_LIB.MXNDArrayCreateSparseEx( - ctypes.c_int(int(_STORAGE_TYPE_STR_TO_ID[stype])), + ctypes.c_int(int(_STORAGE_TYPE_STR_TO_ID[storage_type])), c_array(mx_uint, shape), mx_uint(len(shape)), ctypes.c_int(ctx.device_typeid), @@ -270,7 +270,7 @@ def astype(self, dtype): """ res = mx.nd.zeros(shape=self.shape, ctx=self.context, - dtype=dtype, stype=self.stype) + dtype=dtype, storage_type=self.stype) self.copyto(res) return res @@ -522,7 +522,7 @@ def todense(source): NDArray The dense array with default storage """ - return ndarray.cast_storage(source, stype='default') + return ndarray.cast_storage(source, storage_type='default') def _ndarray_cls(handle, writable=True, stype=None): diff --git a/python/mxnet/symbol.py b/python/mxnet/symbol.py index f6507d8e39e3..796ca77eaa13 100644 --- a/python/mxnet/symbol.py +++ b/python/mxnet/symbol.py @@ -1162,7 +1162,7 @@ def _get_ndarray_inputs(arg_key, args, arg_names, allow_missing): raise TypeError('Only accept list of NDArrays or dict of str to NDArray') return c_array(NDArrayHandle, arg_handles), arg_arrays - def simple_bind(self, ctx, grad_req='write', type_dict=None, stype_dict=None, + def simple_bind(self, ctx, grad_req='write', type_dict=None, storage_type_dict=None, group2ctx=None, shared_arg_names=None, shared_exec=None, shared_buffer=None, **kwargs): """Bind current symbol to get an executor, allocate all the arguments needed. @@ -1206,7 +1206,7 @@ def simple_bind(self, ctx, grad_req='write', type_dict=None, stype_dict=None, type_dict : Dict of str->numpy.dtype Input type dictionary, name->dtype - stype_dict : Dict of str->str + storage_type_dict : Dict of str->str Input storage type dictionary, name->storage_type group2ctx : Dict of string to mx.Context @@ -1255,10 +1255,10 @@ def simple_bind(self, ctx, grad_req='write', type_dict=None, stype_dict=None, # provided storage type argument names provided_arg_stype_names = ctypes.POINTER(ctypes.c_char_p)() provided_arg_stype_data = ctypes.POINTER(mx_uint)() # provided storage types - if stype_dict is not None: + if storage_type_dict is not None: provided_arg_stype_names = [] provided_arg_stype_data = [] - for k, v in stype_dict.items(): + for k, v in storage_type_dict.items(): if v in _STORAGE_TYPE_STR_TO_ID: provided_arg_stype_names.append(c_str(k)) provided_arg_stype_data.append(ctypes.c_int(_STORAGE_TYPE_STR_TO_ID[v])) @@ -1339,7 +1339,7 @@ def simple_bind(self, ctx, grad_req='write', type_dict=None, stype_dict=None, shared_buffer_names = [] shared_buffer_handles = [] for k, v in shared_buffer.items(): - assert(v.stype == 'default'), \ + assert(v.storage_type == 'default'), \ "shared_buffer is expected to only contain NDArrays with default storage" shared_buffer_names.append(c_str(k)) shared_buffer_handles.append(v.handle) @@ -1669,7 +1669,7 @@ def reshape(self, shape): return reshape(self, shape=shape) def var(name, attr=None, shape=None, lr_mult=None, wd_mult=None, dtype=None, - init=None, stype=None, **kwargs): + init=None, storage_type=None, **kwargs): """Creates a symbolic variable with specified name. Example usage: @@ -1696,8 +1696,6 @@ def var(name, attr=None, shape=None, lr_mult=None, wd_mult=None, dtype=None, The dtype for input variable. If not specified, this value will be inferred. init : initializer (mxnet.init.*) Initializer for this variable to (optionally) override the default initializer. - stype : str - The storage type of the variable. kwargs : Additional attribute variables Additional attributes must start and end with double underscores. @@ -1725,8 +1723,8 @@ def var(name, attr=None, shape=None, lr_mult=None, wd_mult=None, dtype=None, if not isinstance(init, string_types): init = init.dumps() attr['__init__'] = init - if stype is not None: - attr['__storage_type__'] = str(_STORAGE_TYPE_STR_TO_ID[stype]) + if storage_type is not None: + attr['__storage_type__'] = str(_STORAGE_TYPE_STR_TO_ID[storage_type]) for k, v in kwargs.items(): if k.startswith('__') and k.endswith('__'): attr[k] = str(v) diff --git a/python/mxnet/test_utils.py b/python/mxnet/test_utils.py index b079d01a85ca..d860b531e520 100644 --- a/python/mxnet/test_utils.py +++ b/python/mxnet/test_utils.py @@ -76,10 +76,10 @@ def random_sample(population, k): return population_copy[0:k] -def rand_sparse_ndarray(shape, stype, density=None): +def rand_sparse_ndarray(shape, storage_type, density=None): """Generate a random sparse ndarray. Returns the ndarray, value(np) and indices(np) """ density = rnd.rand() if density is None else density - if stype == 'row_sparse': + if storage_type == 'row_sparse': # TODO(haibin) support high dim sparse ndarray assert(len(shape) < 3) prod = np.prod(shape) @@ -88,13 +88,13 @@ def rand_sparse_ndarray(shape, stype, density=None): idx_sample = rnd.rand(shape[0]) indices = np.argwhere(idx_sample < density).flatten() if indices.shape[0] == 0: - result = mx.nd.zeros(shape, stype='row_sparse') + result = mx.nd.zeros(shape, storage_type='row_sparse') return result, (np.array([], dtype='int64'), np.array([], dtype='int64')) # generate random values val = rnd.rand(indices.shape[0], num_cols) arr = mx.sparse_nd.row_sparse(val, indices, shape, indices_type=np.int64) return arr, (val, indices) - elif stype == 'csr': + elif storage_type == 'csr': assert(len(shape) == 2) csr = sp.rand(shape[0], shape[1], density=density, format='csr') result = mx.sparse_nd.csr(csr.data, csr.indptr, csr.indices, shape) @@ -103,11 +103,11 @@ def rand_sparse_ndarray(shape, stype, density=None): assert(False), "unknown storage type" -def rand_ndarray(shape, stype, density=None): - if stype == 'default': +def rand_ndarray(shape, storage_type, density=None): + if storage_type == 'default': arr = mx.nd.array(random_arrays(shape)) else: - arr, _ = rand_sparse_ndarray(shape, stype, density=density) + arr, _ = rand_sparse_ndarray(shape, storage_type, density=density) return arr @@ -554,7 +554,7 @@ def random_projection(shape): assert isinstance(grad_stype_dict, dict), "grad_stype_dict must be a dict" for k, v in grad_stype_dict.items(): if k in args_grad and v in _STORAGE_TYPE_STR_TO_ID and v != 'default': - args_grad[k] = mx.nd.cast_storage(args_grad[k], stype=v) + args_grad[k] = mx.nd.cast_storage(args_grad[k], storage_type=v) executor = out.bind(ctx, grad_req=grad_req, args=location, args_grad=args_grad, aux_states=aux_states) @@ -724,7 +724,7 @@ def check_symbolic_backward(sym, location, out_grads, expected, rtol=1e-5, atol= grad_stype = attr.get('grad_stype_hint', None) nd = mx.nd.array(v, ctx=ctx) if grad_stype is not None: - out = mx.nd.cast_storage(nd, stype=grad_stype) + out = mx.nd.cast_storage(nd, storage_type=grad_stype) args_grad_data[k] = out else: args_grad_data[k] = nd diff --git a/src/operator/tensor/cast_storage-inl.h b/src/operator/tensor/cast_storage-inl.h index da9ed30b998a..9273b996d48e 100644 --- a/src/operator/tensor/cast_storage-inl.h +++ b/src/operator/tensor/cast_storage-inl.h @@ -291,9 +291,9 @@ void CastStorageComputeImpl(mshadow::Stream* s, } struct CastStorageParam : public dmlc::Parameter { - int stype; + int storage_type; DMLC_DECLARE_PARAMETER(CastStorageParam) { - DMLC_DECLARE_FIELD(stype) + DMLC_DECLARE_FIELD(storage_type) .add_enum("default", kDefaultStorage) .add_enum("row_sparse", kRowSparseStorage) .add_enum("csr", kCSRStorage) @@ -310,9 +310,9 @@ inline bool CastStorageInferStorageType(const nnvm::NodeAttrs& attrs, CHECK_NE(in_attrs->at(0), kUndefinedStorage) << "src ndarray's storage type must be specified"; const CastStorageParam& param = nnvm::get(attrs.parsed); - CHECK_NE(param.stype, kUndefinedStorage) + CHECK_NE(param.storage_type, kUndefinedStorage) << "dst ndarray's storage type must be specified"; - TYPE_ASSIGN_CHECK(*out_attrs, 0, param.stype); + TYPE_ASSIGN_CHECK(*out_attrs, 0, param.storage_type); return true; } diff --git a/tests/python/unittest/test_kvstore.py b/tests/python/unittest/test_kvstore.py index 1489b8687c26..bd12f95b2496 100644 --- a/tests/python/unittest/test_kvstore.py +++ b/tests/python/unittest/test_kvstore.py @@ -11,9 +11,9 @@ def init_kv(stype='default'): """init kv """ kv = mx.kv.create() # single - kv.init(3, mx.nd.zeros(shape=shape, stype=stype)) + kv.init(3, mx.nd.zeros(shape=shape, storage_type=stype)) # list - kv.init(keys, [mx.nd.zeros(shape=shape, stype=stype)] * len(keys)) + kv.init(keys, [mx.nd.zeros(shape=shape, storage_type=stype)] * len(keys)) return kv def init_kv_with_str(): diff --git a/tests/python/unittest/test_module.py b/tests/python/unittest/test_module.py index 60b06e0f5ce8..96fd77334d8d 100644 --- a/tests/python/unittest/test_module.py +++ b/tests/python/unittest/test_module.py @@ -380,10 +380,10 @@ def test_module_fm(): rnd.seed(11) def fm_model(k, feature_dim): norm = mx.initializer.Normal(sigma=0.01) - x = mx.symbol.Variable("data", stype='csr') - v = mx.symbol.Variable("v", shape=(feature_dim, k), init=norm, stype='row_sparse') + x = mx.symbol.Variable("data", storage_type='csr') + v = mx.symbol.Variable("v", shape=(feature_dim, k), init=norm, storage_type='row_sparse') - w1_weight = mx.symbol.var('w1_weight', shape=(feature_dim, 1), init=norm, stype='row_sparse') + w1_weight = mx.symbol.var('w1_weight', shape=(feature_dim, 1), init=norm, storage_type='row_sparse') w1 = mx.symbol.dot(x, w1_weight) v_s = mx.symbol.sum(data=mx.symbol.square(data=v), axis=1) @@ -443,9 +443,9 @@ def fm_model(k, feature_dim): def test_module_initializer(): def regression_model(m): - x = mx.symbol.var("data", stype='csr') + x = mx.symbol.var("data", storage_type='csr') v = mx.symbol.var("v", shape=(m, 1), init=mx.init.Uniform(scale=.1), - stype='row_sparse') + storage_type='row_sparse') model = mx.symbol.dot(lhs=x, rhs=v) y = mx.symbol.Variable("label") model = mx.symbol.LinearRegressionOutput(data=model, label=y, name="out") @@ -454,7 +454,7 @@ def regression_model(m): n, m = 128, 100 model = regression_model(m) - data = mx.nd.zeros(shape=(n, m), stype='csr') + data = mx.nd.zeros(shape=(n, m), storage_type='csr') label = mx.nd.zeros((n, 1)) iterator = mx.io.NDArrayIter(data=data, label={'label':label}, batch_size=n) diff --git a/tests/python/unittest/test_multi_device_exec.py b/tests/python/unittest/test_multi_device_exec.py index 9823036867d6..3293ae2b0abc 100644 --- a/tests/python/unittest/test_multi_device_exec.py +++ b/tests/python/unittest/test_multi_device_exec.py @@ -35,8 +35,8 @@ def test_ctx_group(): def check_ctx_group_sparse(lhs_stype, rhs_stype): with mx.AttrScope(ctx_group='stage1'): - lhs = mx.symbol.Variable('lhs', stype=lhs_stype) - rhs = mx.symbol.Variable('rhs', stype=rhs_stype) + lhs = mx.symbol.Variable('lhs', storage_type=lhs_stype) + rhs = mx.symbol.Variable('rhs', storage_type=rhs_stype) plus = mx.symbol.elemwise_add(lhs, rhs, name='plus') set_stage1 = set(plus.list_arguments()) diff --git a/tests/python/unittest/test_sparse_ndarray.py b/tests/python/unittest/test_sparse_ndarray.py index f4de352a4c29..66e13801cc30 100644 --- a/tests/python/unittest/test_sparse_ndarray.py +++ b/tests/python/unittest/test_sparse_ndarray.py @@ -15,13 +15,13 @@ def assert_fcompex(f, *args, **kwargs): def sparse_nd_ones(shape, stype): - return mx.nd.cast_storage(mx.nd.ones(shape), stype=stype) + return mx.nd.cast_storage(mx.nd.ones(shape), storage_type=stype) -def check_sparse_nd_elemwise_binary(shapes, stypes, f, g): +def check_sparse_nd_elemwise_binary(shapes, storage_types, f, g): # generate inputs nds = [] - for i, storage_type in enumerate(stypes): + for i, storage_type in enumerate(storage_types): if storage_type == 'row_sparse': nd, _ = rand_sparse_ndarray(shapes[i], storage_type) elif storage_type == 'default': @@ -63,7 +63,7 @@ def test_sparse_nd_elementwise_fallback(): def test_sparse_nd_zeros(): def check_sparse_nd_zeros(stype, shape): zero = mx.nd.zeros(shape) - sparse_zero = mx.nd.zeros(shape=shape, stype=stype) + sparse_zero = mx.nd.zeros(shape=shape, storage_type=stype) assert_almost_equal(sparse_zero.asnumpy(), zero.asnumpy()) shape = rand_shape_2d() @@ -137,8 +137,8 @@ def check_csr_creation(shape): def test_sparse_nd_setitem(): - def check_sparse_nd_setitem(stype, shape, dst): - x = mx.nd.zeros(shape=shape, stype=stype) + def check_sparse_nd_setitem(storage_type, shape, dst): + x = mx.nd.zeros(shape=shape, storage_type=storage_type) x[:] = dst dst_nd = mx.nd.array(dst) if isinstance(dst, (np.ndarray, np.generic)) else dst assert same(x.asnumpy(), dst_nd.asnumpy()) @@ -170,7 +170,7 @@ def check_sparse_nd_csr_slice(shape): def test_sparse_nd_equal(): for stype in ['row_sparse', 'csr']: shape = rand_shape_2d() - x = mx.nd.zeros(shape=shape, stype=stype) + x = mx.nd.zeros(shape=shape, storage_type=stype) y = sparse_nd_ones(shape, stype) z = x == y assert (z.asnumpy() == np.zeros(shape)).all() @@ -181,7 +181,7 @@ def test_sparse_nd_equal(): def test_sparse_nd_not_equal(): for stype in ['row_sparse', 'csr']: shape = rand_shape_2d() - x = mx.nd.zeros(shape=shape, stype=stype) + x = mx.nd.zeros(shape=shape, storage_type=stype) y = sparse_nd_ones(shape, stype) z = x != y assert (z.asnumpy() == np.ones(shape)).all() @@ -192,7 +192,7 @@ def test_sparse_nd_not_equal(): def test_sparse_nd_greater(): for stype in ['row_sparse', 'csr']: shape = rand_shape_2d() - x = mx.nd.zeros(shape=shape, stype=stype) + x = mx.nd.zeros(shape=shape, storage_type=stype) y = sparse_nd_ones(shape, stype) z = x > y assert (z.asnumpy() == np.zeros(shape)).all() @@ -205,7 +205,7 @@ def test_sparse_nd_greater(): def test_sparse_nd_greater_equal(): for stype in ['row_sparse', 'csr']: shape = rand_shape_2d() - x = mx.nd.zeros(shape=shape, stype=stype) + x = mx.nd.zeros(shape=shape, storage_type=stype) y = sparse_nd_ones(shape, stype) z = x >= y assert (z.asnumpy() == np.zeros(shape)).all() @@ -220,7 +220,7 @@ def test_sparse_nd_greater_equal(): def test_sparse_nd_lesser(): for stype in ['row_sparse', 'csr']: shape = rand_shape_2d() - x = mx.nd.zeros(shape=shape, stype=stype) + x = mx.nd.zeros(shape=shape, storage_type=stype) y = sparse_nd_ones(shape, stype) z = y < x assert (z.asnumpy() == np.zeros(shape)).all() @@ -233,7 +233,7 @@ def test_sparse_nd_lesser(): def test_sparse_nd_lesser_equal(): for stype in ['row_sparse', 'csr']: shape = rand_shape_2d() - x = mx.nd.zeros(shape=shape, stype=stype) + x = mx.nd.zeros(shape=shape, storage_type=stype) y = sparse_nd_ones(shape, stype) z = y <= x assert (z.asnumpy() == np.zeros(shape)).all() @@ -328,7 +328,7 @@ def test_sparse_nd_negate(): def test_sparse_nd_output_fallback(): shape = (10, 10) - out = mx.nd.zeros(shape=shape, stype='row_sparse') + out = mx.nd.zeros(shape=shape, storage_type='row_sparse') mx.nd.random_normal(shape=shape, out=out) assert(np.sum(out.asnumpy()) != 0) @@ -336,7 +336,7 @@ def test_sparse_nd_output_fallback(): def test_sparse_nd_astype(): stypes = ['row_sparse', 'csr'] for stype in stypes: - x = mx.nd.zeros(shape=rand_shape_2d(), stype=stype, dtype='float32') + x = mx.nd.zeros(shape=rand_shape_2d(), storage_type=stype, dtype='float32') y = x.astype('int32') assert(y.dtype == np.int32), y.dtype diff --git a/tests/python/unittest/test_sparse_operator.py b/tests/python/unittest/test_sparse_operator.py index bf438f3dd22b..f712da00051d 100644 --- a/tests/python/unittest/test_sparse_operator.py +++ b/tests/python/unittest/test_sparse_operator.py @@ -2,8 +2,8 @@ def check_elemwise_add_ex(lhs_stype, rhs_stype, shape, lhs_grad_stype=None, rhs_grad_stype=None): - lhs = mx.symbol.Variable('lhs', stype=lhs_stype) - rhs = mx.symbol.Variable('rhs', stype=rhs_stype) + lhs = mx.symbol.Variable('lhs', storage_type=lhs_stype) + rhs = mx.symbol.Variable('rhs', storage_type=rhs_stype) if lhs_grad_stype is not None: lhs._set_attr(grad_stype_hint=str(lhs_grad_stype)) if rhs_grad_stype is not None: @@ -48,8 +48,8 @@ def test_elemwise_add_ex_multiple_stages(): ds_nd = mx.nd.array(ds_np) # sparse + sparse = sparse - sp_data1 = mx.symbol.Variable('sp_data1', stype='row_sparse') - sp_data2 = mx.symbol.Variable('sp_data2', stype='row_sparse') + sp_data1 = mx.symbol.Variable('sp_data1', storage_type='row_sparse') + sp_data2 = mx.symbol.Variable('sp_data2', storage_type='row_sparse') ds_data = mx.symbol.Variable('ds_data') plus = mx.symbol.elemwise_add(sp_data1, sp_data2, name='plus') # sparse + dense = dense @@ -69,7 +69,7 @@ def test_elemwise_add_ex_multiple_stages(): def test_cast_storage_ex(): def test_rsp_to_dns(shape): rsp, (data, row_idx) = rand_sparse_ndarray(shape, 'row_sparse') - dns_out = mx.nd.cast_storage(rsp, stype='default') + dns_out = mx.nd.cast_storage(rsp, storage_type='default') dns_expected = np.zeros(shape, dtype=default_dtype()) if row_idx is not None: for k, v in enumerate(row_idx): @@ -78,8 +78,8 @@ def test_rsp_to_dns(shape): def test_dns_to_rsp(shape): dns_in = rand_ndarray(shape, 'default') - rsp_out = mx.nd.cast_storage(mx.nd.array(dns_in, dtype=default_dtype()), stype='row_sparse') - ret = mx.nd.cast_storage(rsp_out, stype='default') + rsp_out = mx.nd.cast_storage(mx.nd.array(dns_in, dtype=default_dtype()), storage_type='row_sparse') + ret = mx.nd.cast_storage(rsp_out, storage_type='default') assert same(ret.asnumpy(), dns_in.asnumpy()) def test_csr_to_dns(shape): @@ -90,8 +90,8 @@ def test_csr_to_dns(shape): def test_dns_to_csr(dns_in): dns_in = np.array(dns_in) - csr_out = mx.nd.cast_storage(mx.nd.array(dns_in, dtype=default_dtype()), stype='csr') - ret = mx.nd.cast_storage(csr_out, stype='default') + csr_out = mx.nd.cast_storage(mx.nd.array(dns_in, dtype=default_dtype()), storage_type='csr') + ret = mx.nd.cast_storage(csr_out, storage_type='default') assert same(ret.asnumpy(), dns_in) shape = rand_shape_2d() @@ -104,7 +104,7 @@ def test_dns_to_csr(dns_in): def test_sparse_dot(): def test_dot_csr(lhs_shape, rhs_shape, rhs_stype, trans_lhs, density=1): lhs_dns = rand_ndarray(lhs_shape, 'default') - lhs_nd = mx.nd.cast_storage(lhs_dns, stype='csr') + lhs_nd = mx.nd.cast_storage(lhs_dns, storage_type='csr') rhs_nd = rand_ndarray(rhs_shape, rhs_stype, density=density) rhs_dns = rhs_nd if rhs_stype == 'default' else rhs_nd.todense() out = mx.nd.dot(lhs_nd, rhs_dns, transpose_a=trans_lhs) @@ -119,8 +119,8 @@ def test_dot_csr(lhs_shape, rhs_shape, rhs_stype, trans_lhs, density=1): assert_almost_equal(out.asnumpy(), out_np, rtol=1e-4, atol=1e-5) # test symbolic forward - lhs = mx.symbol.Variable('lhs', stype='csr') - rhs = mx.symbol.Variable('rhs', stype=rhs_stype) + lhs = mx.symbol.Variable('lhs', storage_type='csr') + rhs = mx.symbol.Variable('rhs', storage_type=rhs_stype) test = mx.symbol.dot(lhs, rhs, transpose_a=trans_lhs) location = {'lhs': lhs_nd, 'rhs': rhs_nd} expected = {'rhs': rhs_backward_grad} @@ -144,7 +144,7 @@ def test_sparse_embedding(): out_dim = 4 batch = 24 - data = mx.sym.Variable("data", stype='csr') + data = mx.sym.Variable("data", storage_type='csr') embed = mx.sym.SparseEmbedding(data=data, input_dim=in_dim, output_dim=out_dim, name="embed") exe_test = embed.simple_bind(default_context(), grad_req={'data': 'null', 'embed_weight': 'write'}, data=(batch, in_dim)) @@ -188,7 +188,7 @@ def test_sparse_retain(): for _ in range(10): shape = rand_shape_2d() num_rows = shape[0] - rsp, _ = rand_sparse_ndarray(shape=shape, stype='row_sparse', density=0.5) + rsp, _ = rand_sparse_ndarray(shape=shape, storage_type='row_sparse', density=0.5) length = np.random.randint(1, num_rows + 1) idx = random_sample(list(range(0, num_rows)), length) idx.sort()