def _py_func(self, hypotheses, references):
"""Wrapper function that converts tensors to unicode and slices
them until the EOS token is found.
"""
# Deal with byte chars
if hypotheses.dtype.kind == np.dtype("U"):
hypotheses = np.char.encode(hypotheses, "utf-8")
if references.dtype.kind == np.dtype("U"):
references = np.char.encode(references, "utf-8")
# Convert back to unicode object
hypotheses = [_.decode("utf-8") for _ in hypotheses]
references = [_.decode("utf-8") for _ in references]
# Slice all hypotheses and references up to SOS -> EOS
sliced_hypotheses = [
postproc.slice_text(_, self._eos_token, self._sos_token)
for _ in hypotheses
]
sliced_references = [
postproc.slice_text(_, self._eos_token, self._sos_token)
for _ in references
]
# Apply postprocessing function
if self._postproc_fn:
sliced_hypotheses = [
self._postproc_fn(_) for _ in sliced_hypotheses
]
sliced_references = [
self._postproc_fn(_) for _ in sliced_references
]
return self.metric_fn(sliced_hypotheses, sliced_references)
def _py_func(self, hypotheses, references):
"""Wrapper function that converts tensors to unicode and slices
them until the EOS token is found.
"""
# Deal with byte chars
if hypotheses.dtype.kind == np.dtype("U"):
hypotheses = np.char.encode(hypotheses, "utf-8")
if references.dtype.kind == np.dtype("U"):
references = np.char.encode(references, "utf-8")
# Convert back to unicode object
hypotheses = [_.decode("utf-8") for _ in hypotheses]
references = [_.decode("utf-8") for _ in references]
# Slice all hypotheses and references up to SOS -> EOS
sliced_hypotheses = [postproc.slice_text(
_, self._eos_token, self._sos_token) for _ in hypotheses]
sliced_references = [postproc.slice_text(
_, self._eos_token, self._sos_token) for _ in references]
# Apply postprocessing function
if self._postproc_fn:
sliced_hypotheses = [self._postproc_fn(_) for _ in sliced_hypotheses]
sliced_references = [self._postproc_fn(_) for _ in sliced_references]
return self.metric_fn(sliced_hypotheses, sliced_references) #pylint: disable=E1102