mindnlp.engine.metrics.perplexity 源代码

# Copyright 2022 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.
# ============================================================================
""""Class for Metric Perplexity"""

import numpy as np
from mindspore import Tensor
from mindnlp.abc import Metric
from mindnlp.common.metrics import _check_value_type, _convert_data_type, _check_onehot_data


[文档]class Perplexity(Metric): r""" Calculates the perplexity. Perplexity is a measure of how well a probabilibity model predicts a sample. A low perplexity indicates the model is good at predicting the sample. The function is shown as follows: .. math:: PP(W)=P(w_{1}w_{2}...w_{N})^{-\frac{1}{N}}=\sqrt[N]{\frac{1}{P(w_{1}w_{2}...w_{N})}} Where :math:`w` represents words in corpus. Args: ignore_label (Union[int, None]): Index of an invalid label to be ignored when counting. If set to `None`, it means there's no invalid label. Default: None. name (str): Name of the metric. Examples: >>> import numpy as np >>> import mindspore >>> from mindspore import Tensor >>> from mindnlp.common.metrics import Perplexity >>> preds = Tensor(np.array([[0.2, 0.5], [0.3, 0.1], [0.9, 0.6]])) >>> labels = Tensor(np.array([1, 0, 1])) >>> metric = Perplexity() >>> metric.update(preds, labels) >>> ppl = metric.eval() >>> print(ppl) 2.231443166940565 """ def __init__(self, ignore_label=None, name='Perplexity'): super().__init__() self._name = name if ignore_label is not None: self.ignore_label = _check_value_type("ignore_label", ignore_label, [int]) else: self.ignore_label = None self.sum_cross_entropy = 0.0 self.sum_word_num = 0
[文档] def clear(self): """Clears the internal evaluation results.""" self.sum_cross_entropy = 0.0 self.sum_word_num = 0
[文档] def update(self, *inputs): """ Updates local variables. Args: inputs: Input `preds` and `labels`. - preds (Union[Tensor, list, np.ndarray]): Predicted value. `preds` is a list of floating numbers in range :math:`[0, 1]` and the shape of `preds` is :math:`(N, C)` in most cases (not strictly), where :math:`N` is the number of cases and :math:`C` is the number of categories. - labels (Union[Tensor, list, np.ndarray]): Ground truth. `labels` must be in one-hot format that shape is :math:`(N, C)`, or can be transformed to one-hot format that shape is :math:`(N,)`. Raises: ValueError: If the number of `inputs` is not 2. RuntimeError: If `preds` and `labels` have different lengths. RuntimeError: If `pred` and `label` have different shapes. """ if len(inputs) != 2: raise ValueError(f'For `Perplexity.update`, it needs 2 inputs (`preds` and `labels`), ' f'but got {len(inputs)}.') preds = inputs[0] labels = inputs[1] preds = _check_value_type("preds", preds, [Tensor, list, np.ndarray]) labels = _check_value_type("labels", labels, [Tensor, list, np.ndarray]) y_pred = [_convert_data_type(preds)] y_true = [_convert_data_type(labels)] if len(y_pred) != len(y_true): raise RuntimeError(f'For `Perplexity.update`, `preds` and `labels` should have ' f'the same length, but got `preds` length {len(y_pred)}, ' f'`labels` length {len(y_true)})') cross_entropy = 0. word_num = 0 for label, pred in zip(y_true, y_pred): if pred.ndim == label.ndim and _check_onehot_data(label): label = label.argmax(axis=1) if label.size != pred.size / pred.shape[-1]: raise RuntimeError(f'For `Perplexity.update`, `preds` and `labels` should have ' f'the same shape, but got `preds` shape {pred.shape}, label ' f'shape {label.shape}.') label = label.reshape((label.size,)) label_expand = label.astype(int) label_expand = np.expand_dims(label_expand, axis=1) first_indices = np.arange(label_expand.shape[0])[:, None] pred = np.squeeze(pred[first_indices, label_expand]) if self.ignore_label is not None: ignore = (label == self.ignore_label).astype(pred.dtype) word_num -= np.sum(ignore) pred = pred * (1 - ignore) + ignore cross_entropy -= np.sum(np.log(np.maximum(1e-10, pred))) word_num += pred.size self.sum_cross_entropy += cross_entropy self.sum_word_num += word_num
[文档] def eval(self): """ Computes and returns the perplexity. Returns: - **ppl** (float) - The computed result. Raises: RuntimeError: If the sample size is 0. """ if self.sum_word_num == 0: raise RuntimeError(f'Perplexity can not be calculated, because the number of ' f'samples is {0}') ppl = np.exp(self.sum_cross_entropy / self.sum_word_num) return ppl
[文档] def get_metric_name(self): """ Return the name of the metric. """ return self._name