!27038 Modify Tensor API Note And Some Error Information.

Merge pull request !27038 from liuyang/tensor_ms

mindspore/common/tensor.py

@@ -80,7 +80,7 @@ class Tensor(Tensor_):
         >>> print(t2.shape)
         ()
         >>> print(t2.dtype)
-        Float64
+        Float32
         >>>
         >>> # initialize a tensor with a tuple
         >>> t3 = Tensor((1, 2))
@@ -125,8 +125,8 @@ class Tensor(Tensor_):
             if isinstance(input_data, np.ndarray) and input_data.dtype not in valid_dtypes and \
                 input_data.dtype.kind != 'U':  # Support dtype np.str_
                 raise TypeError(f"For Tensor, the input_data is a numpy array, "
-                                f"but it's data type: {input_data.dtype} is not in supported list:\
-                                {list(i.__name__ for i in valid_dtypes)}.")
+                                f"but it's data type: {input_data.dtype} is not in supported list: "
+                                f"{list(i.__name__ for i in valid_dtypes)}.")
             if isinstance(input_data, (tuple, list)):
                 if np.array(input_data).dtype not in valid_dtypes:
                     raise TypeError(f"For Tensor, the input_data is {input_data} that contain unsupported element.")
@@ -454,7 +454,7 @@ class Tensor(Tensor_):
             >>> import numpy as np
             >>> from mindspore import Tensor
             >>> x = Tensor(np.array([[1,2,3],[4,5,6]], dtype=np.float32))
-            >>> print(itemset((0,1), 4))
+            >>> print(x.itemset((0,1), 4))
             [[1. 4. 3.]
             [4. 5. 6.]]
             >>> print(x)
@@ -477,7 +477,7 @@ class Tensor(Tensor_):
             >>> import numpy as np
             >>> x = Tensor(np.array([1, 2], dtype=np.float32))
             >>> y = x.asnumpy()
-            >>> x[0] = 100
+            >>> y[0] = 11
             >>> print(x)
             [11.  2.]
             >>> print(y)
@@ -888,15 +888,12 @@ class Tensor(Tensor_):
         Examples:
             >>> import numpy as np
             >>> from mindspore import Tensor
-            >>> x = Tensor(np.ones((1,2,2,1), dtype=np.float32))
+            >>> x = Tensor(np.ones((1,2,2), dtype=np.float32))
             >>> print(x)
-            [[[[1.]
-            [1.]]
-
-            [[1.]
-            [1.]]]]
+            [[[1. 1.]
+            [1. 1.]]]
             >>> print(x.shape)
-            (1, 2, 2, 1)
+            (1, 2, 2)
             >>> y = x.squeeze()
             >>> print(y)
             [[1. 1.]
@@ -905,13 +902,10 @@ class Tensor(Tensor_):
             (2, 2)
             >>> y = x.squeeze(axis=0)
             >>> print(y)
-            [[[1.]
-            [1.]]
-
-            [[1.]
-            [1.]]]
+            [[1. 1.]
+            [1. 1.]]
             >>> print(y.shape)
-            (2, 2, 1)
+            (2, 2)
         """
         self._init_check()
         if axis is None:
@@ -1911,8 +1905,8 @@ class Tensor(Tensor_):
         if not isinstance(ddof, int):
             raise TypeError("For 'Tensor.var', the type of the argument 'ddof' must be int, but got "
                             "{}.".format(type(ddof)))
-        if not isinstance(keepdims, int):
-            raise TypeError("For 'Tensor.var', the type of the argument 'keepdims' must be int, but "
+        if not isinstance(keepdims, bool):
+            raise TypeError("For 'Tensor.var', the type of the argument 'keepdims' must be bool, but "
                             "got {}.".format(type(keepdims)))
 
         if axis is None:

mindspore/context.py

@@ -96,7 +96,7 @@ class _ThreadLocalInfo(threading.local):
     def reserve_class_name_in_scope(self, reserve_class_name_in_scope):
         """Set whether to save the network class name in the scope."""
         if not isinstance(reserve_class_name_in_scope, bool):
-            raise ValueError("For '_ThreadLocalInfo', the type of the property 'reserve_class_name_in_scope' must "
+            raise ValueError("For 'context.set_context', the type of the property 'reserve_class_name_in_scope' must "
                              "be bool, but got {}.".format(type(reserve_class_name_in_scope)))
         self._reserve_class_name_in_scope = reserve_class_name_in_scope
 

mindspore/nn/cell.py

@@ -215,7 +215,7 @@ class Cell(Cell_):
     @cell_init_args.setter
     def cell_init_args(self, value):
         if not isinstance(value, str):
-            raise TypeError(f"For 'Cell', the property 'cell_init_args' must be bool type, "
+            raise TypeError(f"For 'Cell', the property 'cell_init_args' must be string type, "
                             f"but got type {type(value)}.")
         self._cell_init_args = value
 

mindspore/nn/metrics/__init__.py

@@ -120,8 +120,8 @@ def get_metric_fn(name, *args, **kwargs):
         >>> metric = nn.get_metric_fn('precision', eval_type='classification')
     """
     if name not in __factory__:
-        raise KeyError(f"Unsupported metric {name}, please refer to official website for more details about "
-                       f"supported metrics.")
+        raise KeyError(f"For 'get_metric_fn', unsupported metric {name}, please refer to official website "
+                       f"for more details about supported metrics.")
     return __factory__[name](*args, **kwargs)
 
 

mindspore/nn/metrics/accuracy.py

@@ -90,9 +90,9 @@ class Accuracy(EvaluationBase):
         if self._class_num == 0:
             self._class_num = y_pred.shape[1]
         elif y_pred.shape[1] != self._class_num:
-            raise ValueError("Class number not match, last input predicted data contain {} classes, but current "
-                             "predicted data contain {} classes, please check your predicted value(inputs[0])."
-                             .format(self._class_num, y_pred.shape[1]))
+            raise ValueError("For 'Accuracy.update', class number not match, last input predicted data contain {} "
+                             "classes, but current predicted data contain {} classes, please check your predicted "
+                             "value(inputs[0]).".format(self._class_num, y_pred.shape[1]))
 
         if self._type == 'classification':
             indices = y_pred.argmax(axis=1)

mindspore/nn/metrics/bleu_score.py

@@ -146,7 +146,7 @@ class BleuScore(Metric):
 
         """
         if self._is_update is False:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
         if min(self._numerator) == 0.0:
             return np.array(0.0)
 

mindspore/nn/metrics/confusion_matrix.py

@@ -128,7 +128,7 @@ class ConfusionMatrix(Metric):
         """
 
         if not self._is_update:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
 
         confusion_matrix = self.confusion_matrix.astype(float)
 

mindspore/nn/metrics/cosine_similarity.py

@@ -92,7 +92,7 @@ class CosineSimilarity(Metric):
             RuntimeError: If the update method is not called first, an error will be reported.
         """
         if not self._is_update:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
 
         if self.zero_diagonal:
             np.fill_diagonal(self.sqr_mtx_res, 0)

mindspore/nn/metrics/fbeta.py

@@ -49,7 +49,7 @@ class Fbeta(Metric):
         super(Fbeta, self).__init__()
         self.eps = sys.float_info.min
         if not beta > 0:
-            raise ValueError("For Fbeta, the argument 'beta' must be greater than 0, but got {}.".format(beta))
+            raise ValueError("For 'Fbeta', the argument 'beta' must be greater than 0, but got {}.".format(beta))
         self.beta = beta
         self.clear()
 
@@ -83,13 +83,13 @@ class Fbeta(Metric):
         if self._class_num == 0:
             self._class_num = y_pred.shape[1]
         elif y_pred.shape[1] != self._class_num:
-            raise ValueError("Class number not match, last input predicted data contain {} classes, "
-                             "but current predicted data contain {} classes, please check your "
+            raise ValueError("For 'Fbeta.update', class number not match, last input predicted data contain {} "
+                             "classes, but current predicted data contain {} classes, please check your "
                              "predicted value(inputs[0]).".format(self._class_num, y_pred.shape[1]))
         class_num = self._class_num
 
         if y.max() + 1 > class_num:
-            raise ValueError("For 'Fbeta', predicted value(inputs[0]) and true value(inputs[1]) "
+            raise ValueError("For 'Fbeta.update', predicted value(inputs[0]) and true value(inputs[1]) "
                              "should contain same classes, but got predicted value contains {} classes"
                              " and true value contains {} classes.".format(class_num, y.max() + 1))
         y = np.eye(class_num)[y.reshape(-1)]

mindspore/nn/metrics/hausdorff_distance.py

@@ -314,7 +314,7 @@ class HausdorffDistance(Metric):
             RuntimeError: If the update method is not called first, an error will be reported.
         """
         if self._is_update is False:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
 
         hd = self._calculate_percent_hausdorff_distance(self.y_pred_edges, self.y_edges)
         if self.directed:

mindspore/nn/metrics/mean_surface_distance.py

@@ -166,7 +166,7 @@ class MeanSurfaceDistance(Metric):
             RuntimeError: If the update method is not called first, an error will be reported.
         """
         if self._is_update is False:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
 
         mean_surface_distance = self._get_surface_distance(self._y_pred_edges, self._y_edges)
 

mindspore/nn/metrics/metric.py

@@ -94,7 +94,8 @@ class Metric(metaclass=ABCMeta):
         elif isinstance(data, np.ndarray):
             pass
         else:
-            raise TypeError(f'The Input data type must be tensor, list or numpy.ndarray, but got {type(data)}.')
+            raise TypeError(f"For 'Metric' and its derived classes, the input data type must be tensor, list or "
+                            f"numpy.ndarray, but got {type(data)}.")
         return data
 
     def _check_onehot_data(self, data):

mindspore/nn/metrics/occlusion_sensitivity.py

@@ -205,7 +205,7 @@ class OcclusionSensitivity(Metric):
 
         """
         if not self._is_update:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
 
         sensitivity = self._baseline - np.squeeze(self._sensitivity_im)
 

mindspore/nn/metrics/precision.py

@@ -102,9 +102,9 @@ class Precision(EvaluationBase):
         if self._class_num == 0:
             self._class_num = y_pred.shape[1]
         elif y_pred.shape[1] != self._class_num:
-            raise ValueError("Class number not match, last input predicted data contain {} classes, but current "
-                             "predicted data contain {} classes, please check your predicted value(inputs[0])"
-                             .format(self._class_num, y_pred.shape[1]))
+            raise ValueError("For 'Precision.update', class number not match, last input predicted data contain {} "
+                             "classes, but current predicted data contain {} classes, please check your predicted "
+                             "value(inputs[0])".format(self._class_num, y_pred.shape[1]))
 
         class_num = self._class_num
         if self._type == "classification":

mindspore/nn/metrics/recall.py

@@ -103,9 +103,9 @@ class Recall(EvaluationBase):
         if self._class_num == 0:
             self._class_num = y_pred.shape[1]
         elif y_pred.shape[1] != self._class_num:
-            raise ValueError("Class number not match, last input predicted data contain {} classes, but current "
-                             "predicted data contain {} classes, please check your predicted value(inputs[0])."
-                             .format(self._class_num, y_pred.shape[1]))
+            raise ValueError("For 'Recall.update', class number not match, last input predicted data contain {} "
+                             "classes, but current predicted data contain {} classes, please check your predicted "
+                             "value(inputs[0]).".format(self._class_num, y_pred.shape[1]))
 
         class_num = self._class_num
         if self._type == "classification":

mindspore/nn/metrics/roc.py

@@ -206,7 +206,7 @@ class ROC(Metric):
 
         """
         if self._is_update is False:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
 
         y_pred = np.squeeze(np.vstack(self.y_pred))
         y = np.squeeze(np.vstack(self.y))

mindspore/nn/metrics/root_mean_square_surface_distance.py

@@ -165,7 +165,7 @@ class RootMeanSquareDistance(Metric):
 
         """
         if self._is_update is False:
-            raise RuntimeError('Please call the update method before calling eval method.')
+            raise RuntimeError("Please call the 'update' method before calling 'eval' method.")
 
         residual_mean_square_distance = self._get_surface_distance(self._y_pred_edges, self._y_edges)
 

mindspore/nn/optim/optimizer.py

@@ -395,8 +395,8 @@ class Optimizer(Cell):
             return Tensor(np.array(list(learning_rate)).astype(np.float32))
         if isinstance(learning_rate, Tensor):
             if learning_rate.ndim > 1:
-                raise ValueError(f"For 'Optimizer', the dim of Tensor type 'learning_rate' should be a 0 or 1, "
-                                 f"but got {learning_rate.ndim}.")
+                raise ValueError(f"For 'Optimizer', if 'learning_rate' is Tensor type, then the dimension of it should "
+                                 f"be 0 or 1, but got {learning_rate.ndim}.")
             if learning_rate.ndim == 1 and learning_rate.size < 2:
                 logger.warning("If use `Tensor` type dynamic learning rate, please make sure that the number "
                                "of elements in the tensor is greater than 1.")
@@ -523,8 +523,8 @@ class Optimizer(Cell):
             for param in group_param['params']:
                 validator.check_value_type("parameter", param, [Parameter], self.cls_name)
                 if param.name in params_store:
-                    raise RuntimeError(f"The {param.name} parameter already exists, it does not support "
-                                       f"repeated setting. Please check whether the optimizer parameter "
+                    raise RuntimeError(f"For 'Optimizer', the {param.name} parameter already exists, it does not "
+                                       f"support repeated setting. Please check whether the optimizer parameter "
                                        f"has been set multiple times.")
 
                 params_store.append(param.name)
@@ -627,8 +627,8 @@ class Optimizer(Cell):
         for p in param_list:
             validator.check_value_type("parameter", p, [Parameter], self.cls_name)
             if id(p) not in ids:
-                raise ValueError(f"The parameter {p.name} is not in optimizer, please check whether the "
-                                 f"argument 'param' is correct.")
+                raise ValueError(f"For 'get_lr_parameter', the parameter {p.name} is not in optimizer, please check "
+                                 f"whether the argument 'param' is correct.")
             if self.is_group_lr:
                 index = ids.index(id(p))
                 lr.append(get_lr_value(self.learning_rate[index]))
@@ -797,7 +797,7 @@ class _ConvertToCell(LearningRateSchedule):
     def __init__(self, learning_rate):
         super(_ConvertToCell, self).__init__()
         if not isinstance(learning_rate, Parameter):
-            raise TypeError("For '_ConvertToCell', the argument 'learning_rate' must be Parameter, "
+            raise TypeError("For 'Optimizer', the argument 'learning_rate' must be Parameter, "
                             "but got {}.".format(type(learning_rate)))
         self.learning_rate = learning_rate
 
@@ -811,10 +811,10 @@ class _IteratorLearningRate(LearningRateSchedule):
         super(_IteratorLearningRate, self).__init__()
         if isinstance(learning_rate, Tensor):
             if learning_rate.ndim != 1:
-                raise ValueError(f"For '_IteratorLearningRate', the dimension of the argument 'learning_rate' should "
+                raise ValueError(f"For 'Optimizer', the dimension of the argument 'learning_rate' should "
                                  f"be 1, but got {learning_rate.ndim}.")
         else:
-            raise TypeError("For '_IteratorLearningRate', the argument 'learning_rate' should be Tensor, "
+            raise TypeError("For 'Optimizer', the argument 'learning_rate' should be Tensor, "
                             "but got {}.".format(type(learning_rate)))
 
         self.learning_rate = Parameter(learning_rate, name)

mindspore/nn/wrap/cell_wrapper.py

@@ -260,7 +260,7 @@ class ForwardValueAndGrad(Cell):
                             f"the type of 'get_by_list' should be bool, but got '{type(get_by_list)}'")
         if get_by_list and not isinstance(weights, ParameterTuple):
             raise TypeError(f"For 'ForwardValueAndGrad', "
-                            f"when 'get_by_list' is set to True, the argument 'network' should be "
+                            f"when 'get_by_list' is set to True, the argument 'weights' should be "
                             f"ParameterTuple type, but got '{type(weights)}'")
         self.network = network
         if isinstance(network, Cell):

mindspore/nn/wrap/loss_scale.py

@@ -360,7 +360,7 @@ class TrainOneStepWithLossScaleCell(TrainOneStepCell):
             self.scale_sense.set_data(sens)
         else:
             raise TypeError("For 'TrainOneStepWithLossScaleCell', "
-                            "the input type must be Tensor, but got {}".format(type(sens)))
+                            "the type of 'sens' must be Tensor, but got {}".format(type(sens)))
 
     def start_overflow_check(self, pre_cond, compute_input):
         """

mindspore/train/model.py

@@ -65,7 +65,7 @@ def _check_bpckpt_file_name_if_same_exist(directory, prefix):
     suffix_num = 0
     pre_len = len(prefix)
     for filename in files:
-        if filename[-16:] != "breakpoint.ckpt":
+        if filename[-16:] != "_breakpoint.ckpt":
             continue
         # find same prefix file
         if filename.find(prefix) == 0 and not filename[pre_len].isalpha():
@@ -78,7 +78,7 @@ def _check_bpckpt_file_name_if_same_exist(directory, prefix):
                 if num.isdigit():
                     suffix_num = max(suffix_num, int(num)+1)
     if suffix_num != 0:
-        prefix = prefix + "" + str(suffix_num)
+        prefix = prefix + "_" + str(suffix_num)
     return prefix
 
 
@@ -100,7 +100,7 @@ def _save_final_ckpt(func):
             try:
                 func(self, *args, **kwargs)
             except BaseException as e:
-                prefix = self._check_bpckpt_file_name_if_same_exist(obj._directory, obj._exception_prefix)
+                prefix = _check_bpckpt_file_name_if_same_exist(obj._directory, obj._exception_prefix)
                 cur_ckpoint_file = prefix + "-" + str(self._current_epoch_num) + "_" \
                     + str(self._current_step_num) + "_breakpoint.ckpt"
                 cur_file = os.path.join(obj._directory, cur_ckpoint_file)

mindspore/train/serialization.py

@@ -202,7 +202,7 @@ def _exec_save(ckpt_file_name, data_list, enc_key=None, enc_mode="AES-GCM"):
         os.chmod(ckpt_file_name, stat.S_IRUSR)
 
     except BaseException as e:
-        logger.critical("Failed to save the checkpoint file %s. May don't have the permission to write files, "
+        logger.critical("Failed to save the checkpoint file %s. Maybe don't have the permission to write files, "
                         "or the disk space is insufficient and so on.", ckpt_file_name)
         raise e
 
@@ -438,7 +438,7 @@ def load_checkpoint(ckpt_file_name, net=None, strict_load=False, filter_prefix=N
         else:
             pb_content = _decrypt(ckpt_file_name, dec_key, len(dec_key), dec_mode)
             if pb_content is None:
-                raise ValueError('Failed to decrypt the checkpoint file.')
+                raise ValueError("For 'load_checkpoint', Failed to decrypt the checkpoint file.")
         checkpoint_list.ParseFromString(pb_content)
     except BaseException as e:
         if _is_cipher_file(ckpt_file_name):
@@ -447,8 +447,8 @@ def load_checkpoint(ckpt_file_name, net=None, strict_load=False, filter_prefix=N
         else:
             logger.critical("Failed to read the checkpoint file '%s' , may not have permission to read it, please "
                             "check the correct of the file.", ckpt_file_name)
-        raise ValueError(e.__str__() + "\nFailed to read the checkpoint file {}, may not have permission to "
-                                       "read it.".format(ckpt_file_name))
+        raise ValueError(e.__str__() + "\nFor 'load_checkpoint', failed to read the checkpoint file {}, may not have "
+                         "permission to read it.".format(ckpt_file_name))
 
     parameter_dict = {}
     try:
@@ -1413,11 +1413,12 @@ def merge_sliced_parameter(sliced_parameters, strategy=None):
         if parameter.name != parameter_name \
                 or len(parameter.data.shape) != parameter_shape_length \
                 or parameter.data.shape[1:] != parameter_shape[1:]:
-            raise ValueError(f"Please make sure that the elements in 'slice_parameters' have the same name, "
-                             f"dimension length and shape except 0 axis. The name, dimension length, shape except "
-                             f"0 axis should be {parameter_name}, {parameter_shape_length}, {parameter_shape[1:]}, "
-                             f"but got name: {parameter.name}, dimension length: {len(parameter.data.shape)}, "
-                             f"shape except 0 axis: {parameter.data.shape[1:]} at index {index}.")
+            raise ValueError(f"For 'merge_sliced_parameter', please make sure that the elements in 'slice_parameters'"
+                             f" have the same name, dimension length and shape except 0 axis. The name, dimension "
+                             f"length, shape except 0 axis should be {parameter_name}, {parameter_shape_length}, "
+                             f"{parameter_shape[1:]}, but got name: {parameter.name}, dimension length: "
+                             f"{len(parameter.data.shape)}, shape except 0 axis: {parameter.data.shape[1:]} "
+                             f"at index {index}.")
 
         if parameter.data.shape != parameter_shape:
             is_even = False
@@ -1432,8 +1433,8 @@ def merge_sliced_parameter(sliced_parameters, strategy=None):
 
     else:
         if parameter_name not in strategy.keys():
-            raise KeyError(f"The parameter name {parameter_name} should be a key in the 'strategy'. Please check "
-                           f"'sliced_parameter' and 'strategy'.")
+            raise KeyError(f"For 'merge_sliced_parameter', the parameter name {parameter_name} should be a key in "
+                           f"the 'strategy'. Please check 'sliced_parameter' and 'strategy'.")
         merged_tensor = _merge_param_with_strategy(sliced_data, parameter_name, strategy, is_even)
         merged_parameter = Parameter(merged_tensor, parameter_name, requires_grad, layerwise_parallel)
 
@@ -1553,9 +1554,9 @@ def load_distributed_checkpoint(network, checkpoint_filenames, predict_strategy=
             except BaseException as e:
                 logger.critical("Failed to load opt shard slice in load distributed checkpoint for {}. Data shape is {}"
                                 " and group is {}".format(param.name, split_param.data.shape, opt_shard_group))
-                raise RuntimeError(e.__str__() + f"\nFailed to load opt shard slice in load distributed "
-                                                 f"checkpoint for {param.name}. Data shape is {split_param.data.shape} "
-                                                 f"and group is {opt_shard_group}.")
+                raise RuntimeError(e.__str__() + f"\nFor 'load_distributed_checkpoint', failed to load opt shard slice"
+                                   f" in load distributed checkpoint for {param.name}. Data shape is "
+                                   f"{split_param.data.shape} and group is {opt_shard_group}.")
             split_param = Parameter(Tensor(data_slice), param.name,
                                     split_param.requires_grad, split_param.layerwise_parallel)
         param_dict[param.name] = split_param