mindspore 2.1.0__cp38-cp38-manylinux1_x86_64.whl → 2.2.11__cp38-cp38-manylinux1_x86_64.whl

mindspore/safeguard/rewrite_obfuscation.py

@@ -0,0 +1,541 @@
+# Copyright 2023 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""obfuscate network based on rewrite interfaces."""
+import os
+import re
+import secrets
+from pathlib import Path
+
+from mindspore import ops, nn
+from mindspore.common.tensor import Tensor
+from mindspore import log as logger
+from mindspore import load_checkpoint, save_checkpoint
+from mindspore.rewrite import SymbolTree, Node, NodeType, TreeNodeHelper, ScopedValue
+from mindspore.rewrite.parsers.class_def_parser import ClassDefParser
+from mindspore.rewrite.parsers.class_def_parser import ModuleParser
+
+OBF_RATIOS_LENGTH = 1
+MAX_OBF_RATIOS_NUM = 50
+OBF_RATIOS_WIDTH = 0
+OBF_RATIOS_INSERT_INDEX = 0
+
+
+def obfuscate_ckpt(network, ckpt_files, target_modules=None, saved_path='./', obfuscate_scale=100):
+    """
+    obfuscate the plaintext checkpoint files. Usually used in conjunction with
+    :func:`mindspore.load_obf_params_into_net`.
+    interface.
+
+    Args:
+        network (nn.Cell): The original network that need to be obfuscated.
+        ckpt_files (str): The directory path of original ckpt files.
+        target_modules (list[str]): The target module of network that need to be obfuscated. The first string
+            represents the network path of target module in original network, which should be in form of ``'A/B/C'``.
+            The second string represents the obfuscation target module, which should be in form of ``'D|E|F'``. For
+            example, thr target_modules of GPT2 can be ``['backbone/blocks/attention', 'dense1|dense2|dense3']``.
+            If target_modules has the third value, it should be in the format of 'obfuscate_layers:all' or
+            'obfuscate_layers:int', which represents the number of layers need to be obfuscated of duplicate layers
+            (such as transformer layers or resnet blocks). If target_modules is ``None``, the function would search
+            target modules by itself. If found, the searched target module would be used, otherwise suggested target
+            modules would be given with warning log. Default: ``None``.
+        saved_path (str): The directory path for saving obfuscated ckpt files. Default: ``'./'``.
+        obfuscate_scale (Union[float, int]): Obfuscate scale of weights. The generated random obf_ratios will be in
+            range of (1 / obfuscate_scale, obfuscate_scale). Default: 100.
+
+    Raises:
+        TypeError: If `network` is not nn.Cell.
+        TypeError: If `ckpt_files` is not string or `saved_path` is not string.
+        TypeError: If `target_modules` is not list.
+        TypeError: If target_modules's elements are not string.
+        ValueError: If `ckpt_files` is not exist or `saved_path` is not exist.
+        ValueError: If the number of elements of `target_modules` is less than ``2``.
+        ValueError: If the first string of `target_modules` contains characters other than uppercase and lowercase
+            letters, numbers, ``'_'`` and ``'/'``.
+        ValueError: If the second string of `target_modules` is empty or contains characters other than uppercase and
+            lowercase letters, numbers, ``'_'`` and ``'|'``.
+        ValueError: If the third string of `target_modules` is not in the format of 'obfuscate_layers:all' or
+            'obfuscate_layers:int'.
+
+    Returns:
+        list[float], obf_ratios, which is the necessary data that needs to be load when running obfuscated network.
+
+    Examples:
+        >>> from mindspore import obfuscate_ckpt, save_checkpoint
+        >>> # Refer to https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
+        >>> net = LeNet5()
+        >>> save_checkpoint(net, './test_net.ckpt')
+        >>> target_modules = ['', 'fc1|fc2']
+        >>> obfuscate_ckpt(net, target_modules, './', './')
+    """
+    if not isinstance(network, nn.Cell):
+        raise TypeError("network must be nn.Cell, but got {}.".format(type(network)))
+    _check_dir_path('ckpt_files', ckpt_files)
+    _check_dir_path('saved_path', saved_path)
+    # Try to find default target modules
+    if target_modules is None:
+        to_split_modules = _get_default_target_modules(ckpt_files)
+    else:
+        if len(target_modules) >= 1 and target_modules[0] == '/':
+            target_modules[0] = ''
+        to_split_modules = target_modules
+    if not _check_valid_target(network, to_split_modules):
+        raise ValueError("The obfuscate module path {} is not exist, please check the input 'target_modules'."
+                         .format(to_split_modules))
+    if (not isinstance(obfuscate_scale, (float, int))) or (obfuscate_scale <= 1):
+        raise ValueError("obfuscate_scale must be float or int, and larger than 1, but got {}."
+                         .format(obfuscate_scale))
+    # generate and save obf_ratios to saved_path
+    path_list = to_split_modules[0].split('/')
+    target_list = to_split_modules[1].split('|')
+    global OBF_RATIOS_LENGTH
+    number_of_ratios = OBF_RATIOS_LENGTH * OBF_RATIOS_WIDTH
+    if number_of_ratios > MAX_OBF_RATIOS_NUM:
+        OBF_RATIOS_LENGTH = MAX_OBF_RATIOS_NUM // OBF_RATIOS_WIDTH
+        number_of_ratios = OBF_RATIOS_LENGTH * OBF_RATIOS_WIDTH
+    obf_ratios = []
+    secrets_generator = secrets.SystemRandom()
+    for _ in range(number_of_ratios):
+        secure_float = secrets_generator.uniform(1 / obfuscate_scale, obfuscate_scale)
+        obf_ratios.append(secure_float)
+    # start obfuscate ckpt
+    ckpt_dir_files = os.listdir(ckpt_files)
+    for ckpt_name in ckpt_dir_files:
+        sub_path = os.path.abspath(ckpt_files) + '/' + ckpt_name
+        if Path(sub_path).is_dir():
+            sub_ckpt_file_list = os.listdir(sub_path)
+            new_saved_path = os.path.abspath(saved_path) + '/' + ckpt_name
+            if not os.path.exists(new_saved_path):
+                try:
+                    os.mkdir(new_saved_path, mode=0o700)
+                except FileExistsError:
+                    pass
+            for sub_ckpt_name in sub_ckpt_file_list:
+                if not sub_ckpt_name.endswith('.ckpt'):
+                    continue
+                _obfuscate_single_ckpt(os.path.abspath(sub_path) + '/' + sub_ckpt_name, obf_ratios, path_list,
+                                       target_list, new_saved_path)
+        else:
+            if not ckpt_name.endswith('.ckpt'):
+                continue
+            _obfuscate_single_ckpt(os.path.abspath(ckpt_files) + '/' + ckpt_name, obf_ratios, path_list,
+                                   target_list, saved_path)
+    return obf_ratios
+
+
+def _obfuscate_single_ckpt(ckpt_name, obf_ratios, path_list, target_list, saved_path):
+    """Obfuscate single ckpt file"""
+    module_has_been_obfuscated = set()
+    try:
+        ckpt_param = load_checkpoint(ckpt_name)
+    except (ValueError, TypeError, OSError):
+        logger.error("Load checkpoint failed for file {}.".format(ckpt_name))
+        return None
+    obf_ratios_index = -1
+    for item in ckpt_param:
+        module = _get_valid_module(item, path_list, target_list)
+        if module:
+            layer_index = _judge_layer_index(item)
+            if layer_index >= OBF_RATIOS_LENGTH:
+                continue
+            if module not in module_has_been_obfuscated:
+                module_has_been_obfuscated.add(module)
+                obf_ratios_index += 1
+            ratio_total_index = layer_index * OBF_RATIOS_WIDTH + obf_ratios_index % OBF_RATIOS_WIDTH
+            ckpt_param[item].set_data(ckpt_param[item].value() / obf_ratios[ratio_total_index])
+    # save the obfuscated model to saved_path
+    obf_param_list = []
+    for item in ckpt_param:
+        obf_param_list.append({'name': item, 'data': ckpt_param[item]})
+    ckpt_file_name = ckpt_name.split('/')[-1]
+    obf_ckpt_file_name = ckpt_file_name.split('.')[0] + '_obf' + '.ckpt'
+    save_checkpoint(obf_param_list, os.path.abspath(saved_path) + '/' + obf_ckpt_file_name)
+    return None
+
+
+def load_obf_params_into_net(network, target_modules, obf_ratios, data_parallel_num=1, **kwargs):
+    """
+    load obfuscate ratios into obfuscated network. Usually used in conjunction with :func:`mindspore.obfuscate_ckpt`
+    interface.
+
+    Args:
+        network (nn.Cell): The original network that need to be obfuscated.
+        target_modules (list[str]): The target module of network that need to be obfuscated. The first string
+            represents the network path of target module in original network, which should be in form of ``'A/B/C'``.
+            The second string represents the obfuscation target module, which should be in form of ``'D|E|F'``. For
+            example, thr target_modules of GPT2 can be ``['backbone/blocks/attention', 'dense1|dense2|dense3']``.
+            If target_modules has the third value, it should be in the format of 'obfuscate_layers:all' or
+            'obfuscate_layers:int', which represents the number of layers need to be obfuscated of duplicate layers
+            (such as transformer layers or resnet blocks).
+        data_parallel_num (int): The data parallel number of parallel training. Default: 1.
+        obf_ratios (Tensor): The obf ratios generated when execute :func:`mindspore.obfuscate_ckpt`.
+        kwargs (dict): Configuration options dictionary.
+
+            - ignored_func_decorators (list[str]): The name list of function decorators in network's python code.
+            - ignored_class_decorators (list[str]): The name list of class decorators in network's python code.
+
+    Raises:
+        TypeError: If `network` is not nn.Cell.
+        TypeError: If `obf_ratios` is not Tensor.
+        TypeError: If `target_modules` is not list.
+        TypeError: If target_modules's elements are not string.
+        ValueError: If the number of elements of `target_modules` is less than ``2``.
+        ValueError: If `obf_ratios` is empty Tensor.
+        ValueError: If the first string of `target_modules` contains characters other than uppercase and lowercase
+            letters, numbers, ``'_'`` and ``'/'``.
+        ValueError: If the second string of `target_modules` is empty or contains characters other than uppercase and
+            lowercase letters, numbers, ``'_'`` and ``'|'``.
+        ValueError: If the third string of `target_modules` is not in the format of 'obfuscate_layers:all' or
+            'obfuscate_layers:int'.
+        TypeError: If `ignored_func_decorators` is not list[str] or `ignored_class_decorators` is not list[str].
+
+    Examples:
+        >>> from mindspore import obfuscate_ckpt, save_checkpoint, load_checkpoint, Tensor
+        >>> import mindspore.common.dtype as mstype
+        >>> import numpy as np
+        >>> # Refer to https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
+        >>> net = LeNet5()
+        >>> save_checkpoint(net, './test_net.ckpt')
+        >>> target_modules = ['', 'fc1|fc2']
+        >>> # obfuscate ckpt files
+        >>> obfuscate_ckpt(net, target_modules, './', './')
+        >>> # load obf ckpt into network
+        >>> new_net = LeNet5()
+        >>> load_checkpoint('./test_net_obf.ckpt', new_net)
+        >>> obf_ratios = Tensor(np.load('./obf_ratios.npy'), mstype.float16)
+        >>> obf_net = load_obf_params_into_net(new_net, target_modules, obf_ratios)
+    """
+    if not isinstance(network, nn.Cell):
+        raise TypeError("network must be nn.Cell, but got {}.".format(type(network)))
+    if not isinstance(obf_ratios, Tensor):
+        raise TypeError("obf_ratios must be MindSpore Tensor, but got {}.".format(type(obf_ratios)))
+    if obf_ratios.size == 0:
+        raise ValueError("obf_ratios can not be empty.")
+    if not _check_valid_target(network, target_modules):
+        raise ValueError("{} is not exist, please check the input 'target_modules'.".format(target_modules))
+    if (not isinstance(data_parallel_num, int)) or (data_parallel_num <= 0):
+        raise ValueError("data_parallel_num must be positive number, but got {}.".format(data_parallel_num))
+    if len(target_modules) >= 1 and target_modules[0] == '/':
+        target_modules[0] = ''
+    path_list = target_modules[0].split('/')
+    path_len = len(path_list)
+    target_list = []
+    for _ in range(path_len):
+        target_list.append([])
+    target_list.append(target_modules[1].split('|'))
+    global MAX_OBF_RATIOS_NUM, OBF_RATIOS_LENGTH
+    number_of_ratios = OBF_RATIOS_LENGTH * OBF_RATIOS_WIDTH
+    if number_of_ratios > MAX_OBF_RATIOS_NUM:
+        OBF_RATIOS_LENGTH = MAX_OBF_RATIOS_NUM // OBF_RATIOS_WIDTH
+        number_of_ratios = OBF_RATIOS_LENGTH * OBF_RATIOS_WIDTH
+    MAX_OBF_RATIOS_NUM = number_of_ratios
+    rewrite_network = _obfuscate_network(network, path_list, target_list, data_parallel_num=data_parallel_num, **kwargs)
+    setattr(rewrite_network, 'obf_ratios', obf_ratios)
+    return rewrite_network
+
+
+def _check_dir_path(name, dir_path):
+    """check directory path"""
+    if not isinstance(dir_path, str):
+        raise TypeError("{} must be string, but got {}.".format(name, type(dir_path)))
+    if not os.path.exists(dir_path):
+        raise ValueError("{} is not exist, please check the input {}.".format(dir_path, name))
+    if not Path(dir_path).is_dir():
+        raise TypeError("{} must be a directory path, but got {}.".format(name, dir_path))
+
+
+def _judge_layer_index(layer_name):
+    """Judge the layer index of target layers"""
+    split_name = layer_name.split('.')
+    for split_str in split_name[:]:
+        if split_str.isdigit():
+            return int(split_str)
+    return 0
+
+
+def _check_valid_target(network, target_modules):
+    """check whether the input 'target_modules' exists"""
+    if not isinstance(target_modules, list):
+        raise TypeError("target_modules type should be list, but got {}.".format(type(target_modules)))
+    if len(target_modules) < 2:
+        raise ValueError("target_modules should contain at least two string values, in the form of ['A/B/C', 'D1|D2'],"
+                         "but got {}.".format(target_modules))
+    if (not isinstance(target_modules[0], str)) or (not isinstance(target_modules[1], str)):
+        raise TypeError("The values of target_modules should be string, but got {} and {}.".
+                        format(type(target_modules[0]), type(target_modules[1])))
+
+    if not target_modules[1]:
+        raise ValueError("{} should be a non-empty string value, in the form of 'D1|D2'"
+                         .format(target_modules[1]))
+    if not re.fullmatch(pattern=r'([a-zA-Z]*[0-9]*\/*_*)*', string=target_modules[0]) \
+            or not re.fullmatch(pattern=r'([a-zA-Z]*[0-9]*\|*_*)*', string=target_modules[1]):
+        raise ValueError("please check the input 'target_modules'{},it should be in the form of ['A/B/C', 'D1|D2']."
+                         "target_modules[0] can only contain uppercase and lowercase letters, numbers, '_' and '/',"
+                         "target_modules[1] can only contain uppercase and lowercase letters, numbers, '_' and '|'"
+                         .format(target_modules))
+    # target_modules[0] is allowed to be '', it means the main network path
+    path_list = target_modules[0].split('/')
+    target_list = target_modules[1].split('|')
+    net = network
+    # DFS check whether path_list is valid
+    stk = [net]
+    i = 0
+    global OBF_RATIOS_LENGTH
+    OBF_RATIOS_LENGTH = 1
+    while stk and i < len(path_list):
+        net = stk.pop()
+        if hasattr(net, path_list[i]):
+            net = getattr(net, path_list[i])
+            i += 1
+            if isinstance(net, nn.CellList):
+                OBF_RATIOS_LENGTH *= len(net)
+                for n in net:
+                    stk.append(n)
+            elif isinstance(net, nn.Cell):
+                stk.append(net)
+            else:
+                raise TypeError("Target_modules[0] should be a subgraph and it's type should be nn.Cell(nn.CellList),"
+                                "but got type {}".format(type(net)))
+    if target_modules[0] != '' and i != len(path_list):
+        raise ValueError("the path {} does not exist.".format(target_modules[0]))
+    # check whether target_list is valid
+    global OBF_RATIOS_WIDTH
+    OBF_RATIOS_WIDTH = 0
+    for target in target_list:
+        if not hasattr(net, target):
+            logger.warning("{} does not exist in the path {}".format(target, target_modules[0]))
+        else:
+            OBF_RATIOS_WIDTH += 1
+    if OBF_RATIOS_WIDTH == 0:
+        raise ValueError("all targets {} do not exist in the path {}.".format(target_list, target_modules[0]))
+    _update_max_obf_ratios_num(target_modules)
+    return True
+
+
+def _update_max_obf_ratios_num(target_modules):
+    """Update MAX_OBF_RATIOS_NUM"""
+    if len(target_modules) >= 3:
+        obfuscate_layers = target_modules[2].split(':')
+        if len(obfuscate_layers) != 2 or obfuscate_layers[0] != 'obfuscate_layers':
+            raise ValueError("The third value of target_modules should be in the format of 'obfuscate_layers:all' or"
+                             "'obfuscate_layers:int'")
+        global MAX_OBF_RATIOS_NUM
+        if obfuscate_layers[1] == 'all':
+            MAX_OBF_RATIOS_NUM = OBF_RATIOS_LENGTH * OBF_RATIOS_WIDTH
+        else:
+            if not obfuscate_layers[1].isdigit():
+                raise ValueError(
+                    "The third value of target_modules should be in the format of 'obfuscate_layers:all' or"
+                    "'obfuscate_layers:int'")
+            MAX_OBF_RATIOS_NUM = int(obfuscate_layers[1]) * OBF_RATIOS_WIDTH
+
+
+def _get_default_target_modules(ckpt_files):
+    """Get the default or suggested target modules, if the target modules is None."""
+
+    def _split_to_path_and_target(module, target):
+        # split module into path list and target list
+        target_index = module.index(target)
+        path = module[:target_index - 1]
+        target = module[target_index:].split('/')[0]
+        return path, target
+
+    def _find_default_obfuscate_modules(net_path):
+        # find modules including the default paths
+        default_module = {'attention'}
+        for module in default_module:
+            if module in net_path and module not in candidate_modules:
+                candidate_modules.append(net_path)
+        # find the default targets in the default module
+        default_target = {'dense', 'query', 'key', 'value'}
+        for target in default_target:
+            for candidate in candidate_modules:
+                if target in candidate:
+                    path, target = _split_to_path_and_target(candidate, target)
+                    if path not in paths:
+                        paths.append(path)
+                    if target not in targets:
+                        targets.append(target)
+
+    def _find_suggested_obfuscate_modules(net_path):
+        default_target = {'dense', 'query', 'key', 'value'}
+        for target in default_target:
+            # find the suggest modules
+            if target in net_path:
+                path, target = _split_to_path_and_target(net_path, target)
+                if [path, target] not in suggest_modules:
+                    suggest_modules.append([path, target])
+
+    # store the potential candidate_modules
+    candidate_modules = []
+    suggest_modules = []
+    paths = []
+    targets = []
+    ckpt_dir_files = os.listdir(ckpt_files)
+    for ckpt_name in ckpt_dir_files:
+        if not ckpt_name.endswith('.ckpt'):
+            continue
+        try:
+            ckpt_param = load_checkpoint(os.path.abspath(ckpt_files) + '/' + ckpt_name)
+        except (ValueError, TypeError, OSError):
+            logger.error("Load checkpoint failed for file {}.".format(os.path.abspath(ckpt_files) + '/' + ckpt_name))
+            return None
+        for item in ckpt_param:
+            param_path = _remove_digit(item)
+            param_path = '/'.join(param_path)
+            # find candidate modules including the default paths and append candidate_modules
+            _find_default_obfuscate_modules(param_path)
+            # give the suggested modules and find the default targets in the default module
+            _find_suggested_obfuscate_modules(param_path)
+    if paths and targets:
+        target_modules = [paths[0], '|'.join(targets)]
+        logger.warning("The default obfuscate modules is obtained:{}".format(target_modules))
+        return target_modules
+    # logging the suggested target module
+    logger.warning("The default obfuscate modules can not be obtained. The suggested possible paths are given below: {}"
+                   .format(suggest_modules))
+    raise ValueError("Can not get the default path, please specify the path in the form of ['A/B/C', 'D1|D2']")
+
+
+def _get_valid_module(item, path_list, target_list):
+    """get the valid module"""
+    number_path = len(path_list)
+    net_path = _remove_digit(item)
+    net_path = '/'.join(net_path[:number_path])
+    tar_path = '/'.join(path_list)
+    # update the weights with obf_ratios in target module
+    if net_path == tar_path:
+        for target in target_list:
+            if target in item.split('.'):
+                target_index = item.split('.').index(target)
+                module = ''.join(item.split('.')[:target_index + 1])
+                return module
+    return None
+
+
+def _remove_digit(item):
+    """remove digit in the parameter path"""
+    param_path = item.split('.')
+    for tmp_str in param_path[:]:
+        if tmp_str.isdigit():
+            param_path.remove(tmp_str)
+    return param_path
+
+
+def _obfuscate_network(model, path_list, target_list, data_parallel_num=1, **kwargs):
+    """obfuscate original network, including add mul operation and add inputs for passing obf_ratio."""
+
+    def _insert_input(stree: SymbolTree, arg_name: str = 'y_obf'):
+        """add inputs for passing obf_ratio"""
+        last_input = None
+        for node in stree.nodes():
+            if node.get_node_type() == NodeType.Input:
+                last_input = node
+        position = stree.after(last_input)
+        # the insert input node name would be 'input_y_obf'
+        new_input_node = last_input.create_input(arg_name)
+        stree.insert(position, new_input_node)
+
+    def _insert_mul(stree: SymbolTree, node: Node, index: int):
+        """add mul operation for original network"""
+        arg_list = node.get_targets().copy()
+        input_y_node = stree.get_node("input_y_obf")
+        v: str = input_y_node.get_targets()[0].value
+        sv: ScopedValue = ScopedValue.create_naming_value(v + f'[{index}]')
+        arg_list.append(sv)
+        target_list = node.get_targets().copy()
+        if data_parallel_num > 1:
+            logger.info("Data parallel number is: {}".format(data_parallel_num))
+            new_mul_node = node.create_call_cell(cell=ops.Mul().shard(((data_parallel_num, 1), ())),
+                                                 targets=target_list, args=arg_list, name='mul')
+        else:
+            new_mul_node = node.create_call_cell(cell=ops.Mul(), targets=target_list, args=arg_list, name='mul')
+        position = stree.after(node)
+        stree.insert(position, new_mul_node)
+
+    def _insert_mul_by_name(stree: SymbolTree, after_name_list: list):
+        """add mul operation after the target nodes according the name of them"""
+        if not after_name_list:
+            return
+        for node in stree.nodes():
+            for after_name in after_name_list:
+                if node.get_name() == after_name:
+                    global OBF_RATIOS_INSERT_INDEX
+                    if OBF_RATIOS_INSERT_INDEX < MAX_OBF_RATIOS_NUM:
+                        _insert_mul(stree, node, OBF_RATIOS_INSERT_INDEX)
+                        OBF_RATIOS_INSERT_INDEX += 1
+
+    def _update_subnet(stree: SymbolTree, substree: SymbolTree, subnode: Node):
+        """update the network once the subnet is obfuscated"""
+        new_net = substree.get_network()
+        input_y_node = substree.get_node("input_y_obf")
+        if input_y_node is None:
+            return
+        arg_list = subnode.get_args().copy()
+        kwargs_list = list(subnode.get_kwargs().values())
+        arg_list.extend(kwargs_list)
+        v: str = input_y_node.get_targets()[0].value
+        arg_obf: ScopedValue = ScopedValue.create_naming_value("y_obf=" + v)
+        arg_list.append(arg_obf)
+        target_list = subnode.get_targets().copy()
+        name = subnode.get_name()
+        new_node = subnode.create_call_cell(cell=new_net, targets=target_list, args=arg_list, name=name)
+        stree.replace(subnode, [new_node])
+
+    def _traverse(stree, i=0):
+        """traverse and obfuscate the original network"""
+        if len(path_list) == i:
+            return
+        for node in stree.nodes():
+            node_name = node.get_name()
+            if node.get_node_type() == NodeType.Tree and node_name.startswith(path_list[i]):
+                sub_stree = TreeNodeHelper.get_sub_tree(node)
+                _traverse(sub_stree, i + 1)
+                _insert_input(sub_stree, arg_name='y_obf')
+                _insert_mul_by_name(sub_stree, after_name_list=target_list[i + 1])
+                _update_subnet(stree, sub_stree, node)
+
+    def _register_denied_func_decorators(fn):
+        """set the function decorators which should be denied for parse"""
+        name = "denied_function_decorator_list"
+        setattr(ClassDefParser, name, fn)
+
+    def _register_denied_class_decorators(fn):
+        """set the class decorators which should be denied for parse"""
+        name = "denied_class_decorator_list"
+        setattr(ModuleParser, name, fn)
+
+    if 'ignored_func_decorators' in kwargs.keys():
+        kw_func_dec = kwargs["ignored_func_decorators"]
+        if not isinstance(kw_func_dec, list):
+            raise TypeError('{} should be list, but got {}'.format(kw_func_dec, type(kw_func_dec)))
+        if kw_func_dec and not isinstance(kw_func_dec[0], str):
+            raise TypeError('elements of {} should be str, but got {}'.format(kw_func_dec, type(kw_func_dec[0])))
+        _register_denied_func_decorators(kw_func_dec)
+    else:
+        _register_denied_func_decorators(["_args_type_validator_check", "_LogActionOnce", "cell_attr_register"])
+    if 'ignored_class_decorators' in kwargs.keys():
+        kw_class_dec = kwargs["ignored_class_decorators"]
+        _register_denied_class_decorators(kw_class_dec)
+        if not isinstance(kw_class_dec, list):
+            raise TypeError('{} should be list[str] type, but got {}'.format(kw_class_dec, type(kw_class_dec)))
+        if kw_class_dec and not isinstance(kw_class_dec[0], str):
+            raise TypeError('elements of {} should be str, but got {}'.format(kw_class_dec, type(kw_class_dec[0])))
+
+    main_stree = SymbolTree.create(model)
+    _traverse(main_stree, 0)
+    _insert_input(main_stree, arg_name='y_obf')
+    _insert_mul_by_name(main_stree, after_name_list=target_list[0])
+    new_net = main_stree.get_network()
+    return new_net

mindspore/scipy/linalg.py

@@ -461,8 +461,8 @@ def lu_pivots_to_permutation(pivots, permutation_size: int):
         loc = mnp.ix_(*(mnp.arange(0, b) for b in batch_dims))
         x = permutation[..., i]
         y = permutation[loc + (j,)]
-        permutation[..., i] = y
         permutation[loc + (j,)] = x
+        permutation[..., i] = y
     return permutation
 
 

mindspore/scipy/ops.py

@@ -1,4 +1,4 @@
-# Copyright 2021 Huawei Technologies Co., Ltd
+# Copyright 2021-2023 Huawei Technologies Co., Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -156,14 +156,64 @@ class LU(PrimitiveWithInfer):
 
 
 class LinearSumAssignment(Primitive):
-    """Solve the linear sum assignment problem."""
+    r"""
+    Solve the linear sum assignment problem.
+
+    The assignment problem is represented as follows:
+
+    .. math::
+        min\sum_{i}^{} \sum_{j}^{} C_{i,j} X_{i,j}
+
+    where :math:`C` is cost matrix, :math:`X_{i,j} = 1` means column :math:`j` is assigned to row :math:`i` .
+
+    Inputs:
+        - **cost_matrix** (Tensor) - 2-D cost matrix. Tensor of shape :math:`(M, N)` .
+        - **dimension_limit** (Tensor, optional) - A scalar used to limit the actual size of the 2nd dimension of
+          ``cost_matrix``. Default is ``Tensor(sys.maxsize)``, which means no limitation. The type is 0-D int64
+          Tensor.
+        - **maximize** (bool) - Calculate a maximum weight matching if true, otherwise calculate a minimum weight
+          matching.
+
+    Outputs:
+        A tuple of tensors containing 'row_idx' and 'col_idx'.
+
+        - **row_idx** (Tensor) - Row indices of the problem. If `dimension_limit` is given, -1 would be padded at the
+          end. The shape is  :math:`(N, )` , where :math:`N` is the minimum value of `cost_matrix` dimension.
+        - **col_idx** (Tensor) - Column indices of the problem. If `dimension_limit` is given, -1 would be padded at
+          the end. The shape is  :math:`(N, )` , where :math:`N` is the minimum value of `cost_matrix` dimension.
+
+     Raises:
+        TypeError: If the data type of `cost_matrix` is not the type in [float16, float32, float64,
+                   int8, int16, int32, int64, uint8, uint16, uint32, uint64, bool]
+        TypeError: If the type of `maximize` is not bool.
+        TypeError: If the data type of `dimension_limit` is not int64.
+        ValueError: If the rank of `cost_matrix` is not 2.
+        ValueError: If the number of input args is not 3.
+
+
+    Supported Platforms:
+        ``Ascend`` ``CPU``
+
+    Examples:
+        >>> import mindspore as ms
+        >>> import numpy as np
+        >>> from mindspore import Tensor
+        >>> from mindspore.scipy.ops import LinearSumAssignment
+        >>> lsap = LinearSumAssignment()
+        >>> cost_matrix = Tensor(np.array([[2, 3, 3], [3, 2, 3], [3, 3, 2]])).astype(ms.float64)
+        >>> dimension_limit = Tensor(2)
+        >>> maximize = False
+        >>> a, b = lsap(cost_matrix, dimension_limit, maximize)
+        >>> print(a)
+        [0 1 -1]
+        >>> print(b)
+        [0 1 -1]
+    """
 
     @prim_attr_register
     def __init__(self):
-        super().__init__("LinearSumAssignment")
+        super().__init__(name="LinearSumAssignment")
         self.init_prim_io_names(inputs=['cost_matrix', 'dimension_limit', 'maximize'], outputs=['row_ind', 'col_ind'])
-        self.add_prim_attr("cust_aicpu", "mindspore_aicpu_kernels")
-
 
 # pylint: disable=C0413,W0611
 from .ops_grad import get_bprpo_eigh, get_bprpo_trsm

mindspore/scipy/optimize/__init__.py

@@ -1,4 +1,4 @@
-# Copyright 2021 Huawei Technologies Co., Ltd
+# Copyright 2021-2023 Huawei Technologies Co., Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -15,5 +15,6 @@
 """Optimize submodule"""
 from .minimize import minimize
 from .line_search import line_search
+from .linear_sum_assignment import linear_sum_assignment
 
-__all__ = ["minimize", "line_search"]
+__all__ = ["minimize", "line_search", "linear_sum_assignment"]