mindspore 2.1.0__cp37-none-any.whl → 2.2.10__cp37-none-any.whl

mindspore/train/model.py

@@ -1,4 +1,4 @@
-# Copyright 2020-2022 Huawei Technologies Co., Ltd
+# Copyright 2020-2023 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.
@@ -25,6 +25,7 @@ import importlib
 import numpy as np
 
 import mindspore
+import mindspore.dataset as ds
 from mindspore import log as logger
 from mindspore.train.serialization import save_checkpoint, load_checkpoint
 from mindspore.train.callback._checkpoint import ModelCheckpoint, _chg_ckpt_file_name_if_same_exist
@@ -188,7 +189,7 @@ class Model:
         >>> from mindspore.train import Model
         >>>
         >>> # Define the network structure of LeNet5. Refer to
-        >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/lenet.py
+        >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
         >>> net = LeNet5()
         >>> loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
         >>> optim = nn.Momentum(params=net.trainable_params(), learning_rate=0.1, momentum=0.9)
@@ -197,7 +198,7 @@ class Model:
         >>> model.predict_network
         >>> model.eval_network
         >>> # Create the dataset taking MNIST as an example. Refer to
-        >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/mnist.py
+        >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
         >>> dataset = create_dataset()
         >>> model.train(2, dataset)
     """
@@ -224,6 +225,7 @@ class Model:
         self._check_for_graph_cell(kwargs)
         self._build_boost_network(kwargs)
         self._train_network = self._build_train_network()
+        self._train_network._jit_config_dict = network.jit_config_dict
         self._build_eval_network(metrics, self._eval_network, eval_indexes)
         self._build_predict_network()
         self._current_epoch_num = 0
@@ -232,9 +234,12 @@ class Model:
         self.enable_recovery = False
         self._backbone_is_train = True
         self.need_load_ckpt = False
-        self._lite_predictor = None
+        self._lite_full_predictor = None
+        self._lite_incremental_predictor = None
         self._mindspore_lite = None
         self._lite_infer = True  # if backend lite infer fails, set False
+        self._mindspore_lite_model_group_id = id(self) & 0xFFFF
+
 
     def _check_for_graph_cell(self, kwargs):
         """Check for graph cell"""
@@ -575,7 +580,6 @@ class Model:
             initial_epoch (int): Epoch at which to start train, it used for resuming a previous training run.
                                  Default: 0.
         """
-        epoch = Validator.check_positive_int(epoch)
         if self._parameter_broadcast:
             self._train_network.set_broadcast_flag()
 
@@ -648,10 +652,10 @@ class Model:
                            "it is suggested to pad/drop data or adjust sink_size. "
                            "But got 'dataset_size': {}, 'sink_size': {}.".format(dataset_size, sink_size))
         if sink_size == -1:
-            dataset_sink_num = epoch - initial_epoch
+            dataset_sink_num = epoch
         else:
-            dataset_sink_num = math.ceil((epoch - initial_epoch) * sink_size / dataset_size)
-            train_dataset.__total_batch__ = (epoch - initial_epoch) * sink_size
+            dataset_sink_num = math.ceil(epoch * sink_size / dataset_size)
+            train_dataset.__total_batch__ = epoch * sink_size
 
         cb_params.cur_step_num = 0
         cb_params.dataset_sink_mode = True
@@ -684,6 +688,7 @@ class Model:
                                                                   dataset_helper=dataset_helper)
 
             cb_params.train_network = train_network
+            cb_params.dataset_helper = dataset_helper
 
             # Perform recovery for process which is restarted.
             self._reset_training_step_for_abnormal_process(cb_params, dataset_helper)
@@ -829,7 +834,7 @@ class Model:
                 os.remove(cb_params.latest_ckpt_file)
                 raise RuntimeError(e.__str__() + ", load ckpt failed and remove the ckpt: "\
                                    + cb_params.latest_ckpt_file) from e
-            _reset_training_dataset(cb_params.cur_step_num, dataset_helper.sink_size())
+            _reset_training_dataset(cb_params.cur_step_num, dataset_helper.iter.dataset.get_dataset_size())
             self.need_load_ckpt = False
 
     def _reset_training_step_for_normal_process(self, cb_params, dataset_helper):
@@ -858,9 +863,9 @@ class Model:
                 self.epoch_iter = recovery_epoch_num
                 cb_params.cur_epoch_num = self.epoch_iter + 1
                 cb_params.last_save_ckpt_step = cb_params.cur_step_num
-                _reset_training_dataset(cb_params.cur_step_num, dataset_helper.sink_size())
+                _reset_training_dataset(cb_params.cur_step_num, dataset_helper.iter.dataset.get_dataset_size())
             else:
-                _reset_training_dataset(0, dataset_helper.sink_size())
+                _reset_training_dataset(0, dataset_helper.iter.dataset.get_dataset_size())
 
             _set_recovery_context(need_reset=False)
 
@@ -884,7 +889,7 @@ class Model:
         dataset_helper, _ = self._exec_preprocess(is_train=True,
                                                   dataset=train_dataset,
                                                   dataset_sink_mode=False,
-                                                  epoch_num=(epoch-initial_epoch))
+                                                  epoch_num=epoch)
         cb_params.cur_step_num = 0
         cb_params.dataset_sink_mode = False
         run_context = RunContext(cb_params)
@@ -962,7 +967,7 @@ class Model:
             of data will be transferred one by one. The limitation of data transmission per time is 256M.
 
             When dataset_sink_mode is True, the `step_end` method of the instance of Callback will be called at the end
-            of epoch.
+            of step in PyNative mode， or will be called at the end of epoch in Graph mode.
 
             If dataset_sink_mode is True, dataset will be bound to this model and cannot be used by other models.
 
@@ -1003,10 +1008,10 @@ class Model:
             >>> from mindspore.train import Model
             >>>
             >>> # Create the dataset taking MNIST as an example. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/mnist.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
             >>> dataset = create_dataset()
             >>> # Define the network structure of LeNet5. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/lenet.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
             >>> net = LeNet5()
             >>> loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
             >>> loss_scale_manager = ms.FixedLossScaleManager(1024., False)
@@ -1015,6 +1020,8 @@ class Model:
             ...                  loss_scale_manager=loss_scale_manager)
             >>> model.train(2, dataset)
         """
+        # prepare dataset for obfuscated model
+        train_dataset = self._prepare_obf_dataset(train_dataset)
         device_target = context.get_context("device_target")
         if _is_ps_mode() and not _cache_enable() and (device_target in ["Ascend", "CPU"]) and dataset_sink_mode:
             logger.info("For PS mode, reset datasink mode to False when using Ascend or CPU backend.")
@@ -1038,7 +1045,7 @@ class Model:
         self._check_sink_mode_for_ds_debug_mode(dataset_sink_mode)
 
         Validator.check_is_int(sink_size)
-        Validator.check_non_negative_int(epoch)
+        Validator.check_positive_int(epoch)
         Validator.check_non_negative_int(initial_epoch)
         if initial_epoch >= epoch:
             raise ValueError(f"For 'Model.train', the parameter 'epoch' must bigger than parameter 'initial_epoch',"
@@ -1154,11 +1161,11 @@ class Model:
             >>> from mindspore.train import Model
             >>>
             >>> # Create the dataset taking MNIST as an example. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/mnist.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
             >>> train_dataset = create_dataset("train")
             >>> valid_dataset = create_dataset("test")
             >>> # Define the network structure of LeNet5. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/lenet.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
             >>> net = LeNet5()
             >>> loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
             >>> optim = nn.Momentum(params=net.trainable_params(), learning_rate=0.1, momentum=0.9)
@@ -1167,7 +1174,7 @@ class Model:
 
         Tutorial Examples:
             - `Advanced Encapsulation: Model - Train and Save Model
-              <https://mindspore.cn/tutorials/en/r2.1/advanced/model.html#train-and-save-model>`_
+              <https://www.mindspore.cn/tutorials/en/r2.2/advanced/model.html#training-and-saving-model>`_
         """
         device_target = context.get_context("device_target")
         if _is_ps_mode() and not _cache_enable() and (device_target in ["Ascend", "CPU"]) and dataset_sink_mode:
@@ -1186,7 +1193,7 @@ class Model:
                              .format(train_dataset._warmup_epoch, epoch))
 
         Validator.check_is_int(sink_size)
-        Validator.check_non_negative_int(epoch)
+        Validator.check_positive_int(epoch)
         Validator.check_non_negative_int(initial_epoch)
         if initial_epoch >= epoch:
             raise ValueError(f"For 'Model.fit', the parameter 'epoch' must bigger than parameter 'initial_epoch',"
@@ -1247,10 +1254,10 @@ class Model:
             >>> from mindspore.amp import FixedLossScaleManager
             >>>
             >>> # Create the dataset taking MNIST as an example. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/mnist.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
             >>> dataset = create_dataset()
             >>> # Define the network structure of LeNet5. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/lenet.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
             >>> net = LeNet5()
             >>> loss = nn.SoftmaxCrossEntropyWithLogits()
             >>> loss_scale_manager = FixedLossScaleManager()
@@ -1260,6 +1267,10 @@ class Model:
             >>> model.build(dataset, epoch=2)
             >>> model.train(2, dataset)
         """
+        epoch = Validator.check_positive_int(epoch)
+        if hasattr(self._train_network, '_is_check_and_refresh') and not self._train_network._is_check_and_refresh:
+            self._train_network.check_names_and_refresh_name()
+            self._train_network._is_check_and_refresh = True
         self._init(train_dataset, valid_dataset, sink_size, epoch)
 
     def _eval_in_fit(self, valid_dataset, callbacks=None, dataset_sink_mode=True, cb_params=None):
@@ -1419,10 +1430,10 @@ class Model:
             >>> from mindspore.train import Model
             >>>
             >>> # Create the dataset taking MNIST as an example. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/mnist.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
             >>> dataset = create_dataset()
             >>> # Define the network structure of LeNet5. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/lenet.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
             >>> net = LeNet5()
             >>> loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
             >>> model = Model(net, loss_fn=loss, optimizer=None, metrics={'acc'})
@@ -1430,8 +1441,9 @@ class Model:
 
         Tutorial Examples:
             - `Advanced Encapsulation: Model - Train and Save Model
-              <https://mindspore.cn/tutorials/en/r2.1/advanced/model.html#train-and-save-model>`_
+              <https://www.mindspore.cn/tutorials/en/r2.2/advanced/model.html#training-and-saving-model>`_
         """
+        valid_dataset = self._prepare_obf_dataset(valid_dataset)
         dataset_sink_mode = Validator.check_bool(dataset_sink_mode)
 
         _device_number_check(self._parallel_mode, self._device_number)
@@ -1479,7 +1491,7 @@ class Model:
 
         return eval_result
 
-    def _predict_lite(self, *predict_data):
+    def _predict_lite(self, *predict_data, config=None):
         """
         Generate output predictions for the input samples using backend 'lite'.
 
@@ -1488,6 +1500,48 @@ class Model:
                 The predict data, can be a single tensor,
                 a list of tensor, or a tuple of tensor.
 
+            config (dict, optional) - The config parameter is enabled when the backend is ‘lite’.
+                The config includes two parts: config_path (configPath, str) and config_item (str, dict).
+                When the config_item is set, its priority is higher than the config_path. Set the ranking
+                table file for inference. The content of the configuration file is as follows:
+
+                config_path defines the path of the configuration file, which is used to pass user-defined
+                    options during model building. In the following scenarios, users may need to set parameters.
+                    For example: "/home/user/config.ini". Default value: ``"" `` , here is the content of the
+                    config.ini file:
+
+                .. code-block::
+
+                    [ascend_context]
+                    rank_table_file = [path_a](storage initial path of the rank table file)
+                    [execution_plan]
+                    [op_name1] = data_type:float16 (operator named op_name1 is set to data type Float16)
+                    [op_name2] = data_type:float32 (operator named op_name2 is set to data type Float32)
+
+                When only the config_path is configured, it is done as follows:
+
+                .. code-block::
+
+                    config = {"configPath" : "/home/user/config.ini"}
+
+                When only the config_dict is configured, it is done as follows:
+
+                .. code-block::
+
+                    config = {"ascend_context" : {"rank_table_file" : "path_b"},
+                              "execution_plan" : {"op_name1" : "data_type:float16", "op_name2" : "data_type:float32"}}
+
+                When both the `config_path` and the `config_dict` are configured, it is done as follows:
+
+                .. code-block::
+
+                    config = {"configPath" : "/home/user/config.ini",
+                              "ascend_context" : {"rank_table_file" : "path_b"},
+                              "execution_plan" : {"op_name3" : "data_type:float16", "op_name4" : "data_type:float32"}}
+
+                Note that both the "configPath" is configured in the config_dict and the config_item,
+                    in this case, the path_b in the config_dict takes precedence.
+
         Returns:
             Tensor, array(s) of predictions.
         """
@@ -1503,32 +1557,66 @@ class Model:
                 device_id = context.get_context('device_id')
                 if device_id and isinstance(device_id, int):
                     lite_context_input.gpu.device_id = device_id
+                if context.get_auto_parallel_context("parallel_mode") == context.ParallelMode.SEMI_AUTO_PARALLEL:
+                    from mindspore.communication import init, get_rank
+                    init()
+                    lite_context_input.gpu.rank_id = get_rank()
             elif device_target == 'ascend':
                 device_id = context.get_context('device_id')
                 if device_id and isinstance(device_id, int):
                     lite_context_input.ascend.device_id = device_id
+                if context.get_auto_parallel_context("parallel_mode") == context.ParallelMode.SEMI_AUTO_PARALLEL:
+                    from mindspore.communication import init, get_rank
+                    init()
+                    lite_context_input.ascend.rank_id = get_rank()
+                    lite_context_input.ascend.provider = "ge"
             else:
                 raise RuntimeError(f"For predict lite, device target should be in ['gpu', 'cpu', 'ascend']"
                                    f" but got {device_target}")
-
             return lite_context_input
 
         if not self._mindspore_lite:
             self._mindspore_lite = importlib.import_module('mindspore_lite')
 
-        check_input_data(*predict_data, data_class=Tensor)
-        if not self._lite_predictor:
-            lite_context = _get_lite_context(self._mindspore_lite.Context())
-            self._lite_predictor = \
-                self._mindspore_lite.lite_infer.LiteInfer(self, *predict_data, context=lite_context)
-
-        inputs = self._lite_predictor.get_inputs()
-        if len(predict_data) != len(inputs):
-            raise RuntimeError(f"For 'Model.predict', numbers of predict_data {len(predict_data)} "
-                               f"is not equal to numbers of net input {len(inputs)}")
-        for i, single_data in enumerate(predict_data):
-            inputs[i].set_data_from_numpy(single_data.asnumpy())
-        outputs: list = self._lite_predictor.predict(inputs)
+        use_past = False    # default execute full model inference
+        model_group_id = None
+        if self._predict_network.get_flags().__contains__("is_first_iteration"):
+            is_first_iteration = self._predict_network.get_flags()['is_first_iteration']
+            if isinstance(is_first_iteration, bool):
+                use_past = not is_first_iteration
+                model_group_id = self._mindspore_lite_model_group_id
+
+        check_input_data(*predict_data, data_class=(int, float, str, None, Tensor))
+        if use_past:
+            # Execute incremental model inference
+            if not self._lite_incremental_predictor:
+                lite_context = _get_lite_context(self._mindspore_lite.Context())
+                self._lite_incremental_predictor = \
+                    self._mindspore_lite.lite_infer.LiteInfer(self, *predict_data, context=lite_context,
+                                                              model_group_id=model_group_id, config=config)
+
+            inputs = self._lite_incremental_predictor.get_inputs()
+            if len(predict_data) != len(inputs):
+                raise RuntimeError(f"For 'Model.predict', numbers of predict_data {len(predict_data)} "
+                                   f"is not equal to numbers of net input {len(inputs)}")
+            for i, single_data in enumerate(predict_data):
+                inputs[i].set_data_from_numpy(single_data.asnumpy())
+            outputs: list = self._lite_incremental_predictor.predict(inputs)
+        else:
+            # Execute full model inference
+            if not self._lite_full_predictor:
+                lite_context = _get_lite_context(self._mindspore_lite.Context())
+                self._lite_full_predictor = \
+                    self._mindspore_lite.lite_infer.LiteInfer(self, *predict_data, context=lite_context,
+                                                              model_group_id=model_group_id, config=config)
+
+            inputs = self._lite_full_predictor.get_inputs()
+            if len(predict_data) != len(inputs):
+                raise RuntimeError(f"For 'Model.predict', numbers of predict_data {len(predict_data)} "
+                                   f"is not equal to numbers of net input {len(inputs)}")
+            for i, single_data in enumerate(predict_data):
+                inputs[i].set_data_from_numpy(single_data.asnumpy())
+            outputs: list = self._lite_full_predictor.predict(inputs)
         if not outputs:
             return Tensor(outputs)
         if len(outputs) == 1:
@@ -1536,7 +1624,7 @@ class Model:
         outputs = [Tensor(single_output.get_data_to_numpy()) for single_output in outputs]
         return tuple(outputs)
 
-    def predict(self, *predict_data, backend=None):
+    def predict(self, *predict_data, backend=None, config=None):
         """
         Generate output predictions for the input samples.
 
@@ -1545,7 +1633,48 @@ class Model:
                 The predict data, can be a single tensor,
                 a list of tensor, or a tuple of tensor.
             backend (str): Select predict backend, this parameter is an experimental feature
-                and is mainly used for MindSpore Lite cloud-side inference. Default: ``False`` .
+                and is mainly used for MindSpore Lite cloud-side inference. Default: ``None`` .
+            config (dict, optional) - The config parameter is enabled when the backend is ‘lite’.
+                The config includes two parts: config_path (configPath, str) and config_item (str, dict).
+                When the config_item is set, its priority is higher than the config_path. Set the ranking
+                table file for inference. The content of the configuration file is as follows:
+
+                config_path defines the path of the configuration file, which is used to pass user-defined
+                options during model building. In the following scenarios, users may need to set parameters.
+                For example: "/home/user/config.ini". Default value: ``""`` , here is the content of the
+                config.ini file:
+
+                .. code-block::
+
+                    [ascend_context]
+                    rank_table_file = [path_a](storage initial path of the rank table file)
+                    [execution_plan]
+                    [op_name1] = data_type:float16 (operator named op_name1 is set to data type Float16)
+                    [op_name2] = data_type:float32 (operator named op_name2 is set to data type Float32)
+
+                When only the config_path is configured, it is done as follows:
+
+                .. code-block::
+
+                    config = {"configPath" : "/home/user/config.ini"}
+
+                When only the config_dict is configured, it is done as follows:
+
+                .. code-block::
+
+                    config = {"ascend_context" : {"rank_table_file" : "path_b"},
+                              "execution_plan" : {"op_name1" : "data_type:float16", "op_name2" : "data_type:float32"}}
+
+                When both the `config_path` and the `config_dict` are configured, it is done as follows:
+
+                .. code-block::
+
+                    config = {"configPath" : "/home/user/config.ini",
+                              "ascend_context" : {"rank_table_file" : "path_b"},
+                              "execution_plan" : {"op_name3" : "data_type:float16", "op_name4" : "data_type:float32"}}
+
+                Note that both the "configPath" is configured in the config_dict and the config_item,
+                    in this case, the path_b in the config_dict takes precedence.
 
         Returns:
             Tensor, array(s) of predictions.
@@ -1558,7 +1687,7 @@ class Model:
             >>>
             >>> input_data = Tensor(np.random.randint(0, 255, [1, 1, 32, 32]), mindspore.float32)
             >>> # Define the network structure of LeNet5. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/lenet.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
             >>> model = Model(LeNet5())
             >>> result = model.predict(input_data)
         """
@@ -1567,7 +1696,7 @@ class Model:
         if backend == "lite" and self._lite_infer:
             # pylint: disable=broad-except
             try:
-                return self._predict_lite(*predict_data)
+                return self._predict_lite(*predict_data, config=config)
             except RuntimeError:
                 self._lite_infer = False
                 logger.warning("Lite inference failed, fallback to original inference!")
@@ -1666,10 +1795,10 @@ class Model:
             >>> ms.set_auto_parallel_context(parallel_mode=ms.ParallelMode.SEMI_AUTO_PARALLEL)
             >>>
             >>> # Create the dataset taking MNIST as an example. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/mnist.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
             >>> dataset = create_dataset()
             >>> # Define the network structure of LeNet5. Refer to
-            >>> # https://gitee.com/mindspore/docs/blob/r2.1/docs/mindspore/code/lenet.py
+            >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
             >>> net = LeNet5()
             >>> loss = nn.SoftmaxCrossEntropyWithLogits()
             >>> loss_scale_manager = ms.FixedLossScaleManager()
@@ -1738,7 +1867,7 @@ class Model:
         if _get_parallel_mode() not in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL):
             raise RuntimeError('Infer predict layout only supports semi auto parallel and auto parallel mode.')
         _parallel_predict_check()
-        check_input_data(*predict_data, data_class=Tensor)
+        check_input_data(*predict_data, data_class=(int, float, str, None, Tensor))
 
         predict_net = self._predict_network
         # Unlike the cases in build_train_network() and build_eval_network(), 'multi_subgraphs' is not set
@@ -1790,5 +1919,16 @@ class Model:
         """
         return self._eval_network
 
+    def _prepare_obf_dataset(self, dataset):
+        if not hasattr(self._network, 'obf_ratios'):
+            return dataset
+        data_size = dataset.get_dataset_size()
+        obf_ratio_dataset = []
+        for _ in range(data_size):
+            obf_ratio_dataset.append(self._network.obf_ratios)
+        obf_ratio_dataset = ds.NumpySlicesDataset(data=obf_ratio_dataset, column_names=["y_obf"])
+        dataset = ds.zip((dataset, obf_ratio_dataset))
+        return dataset
+
 
 __all__ = ["Model"]