困惑度指标
Perplexity class
keras_nlp.metrics.Perplexity(
from_logits=False, mask_token_id=None, dtype="float32", name="perplexity", **kwargs
)
Perplexity metric.
This class implements the perplexity metric. In short, this class calculates the cross entropy loss and takes its exponent. Note: This implementation is not suitable for fixed-size windows.
Arguments
- from_logits: bool. If True,
y_pred(input toupdate_state()) should be the logits as returned by the model. Otherwise,y_predis a tensor of probabilities. - mask_token_id: int. ID of the token to be masked. If provided, the mask
is computed for this class. Note that if this field is provided, and
if the
sample_weightfield inupdate_state()is also provided, we will compute the finalsample_weightas the element-wise product of the mask and thesample_weight. - dtype: string or tf.dtypes.Dtype. Precision of metric computation. If
not specified, it defaults to
"float32". - name: string. Name of the metric instance.
- **kwargs: Other keyword arguments.
Examples
- Calculate perplexity by calling update_state() and result().
1.1.
sample_weight, andmask_token_idare not provided.
>>> np.random.seed(42)
>>> perplexity = keras_nlp.metrics.Perplexity(name="perplexity")
>>> target = np.random.randint(10, size=[2, 5])
>>> logits = np.random.uniform(size=(2, 5, 10))
>>> perplexity.update_state(target, logits)
>>> perplexity.result()
<tf.Tensor: shape=(), dtype=float32, numpy=14.352535>
1.2. sample_weight specified (masking token with ID 0).
>>> np.random.seed(42)
>>> perplexity = keras_nlp.metrics.Perplexity(name="perplexity")
>>> target = np.random.randint(10, size=[2, 5])
>>> logits = np.random.uniform(size=(2, 5, 10))
>>> sample_weight = (target != 0).astype("float32")
>>> perplexity.update_state(target, logits, sample_weight)
>>> perplexity.result()
<tf.Tensor: shape=(), dtype=float32, numpy=14.352535>
- Call perplexity directly.
>>> np.random.seed(42)
>>> perplexity = keras_nlp.metrics.Perplexity(name="perplexity")
>>> target = np.random.randint(10, size=[2, 5])
>>> logits = np.random.uniform(size=(2, 5, 10))
>>> perplexity(target, logits)
<tf.Tensor: shape=(), dtype=float32, numpy=14.352535>
- Provide the padding token ID and let the class compute the mask on its own.
>>> np.random.seed(42)
>>> perplexity = keras_nlp.metrics.Perplexity(mask_token_id=0)
>>> target = np.random.randint(10, size=[2, 5])
>>> logits = np.random.uniform(size=(2, 5, 10))
>>> perplexity(target, logits)
<tf.Tensor: shape=(), dtype=float32, numpy=14.352535>