def remap(self, field_map: Dict[str, str]):
"""Return a copy of this dataset with some fields renamed."""
new_spec = utils.remap_dict(self.spec(), field_map)
new_examples = [
utils.remap_dict(ex, field_map) for ex in self.examples
]
return Dataset(new_spec, new_examples, base=self)
def test_remap_dict(self):
d = {
"a": True,
"b": False,
"c": True
}
remap_dict = {
"a": "a2",
"b": "b2"
}
expected = {
"a2": True,
"b2": False,
"c": True
}
self.assertDictEqual(expected, utils.remap_dict(d, remap_dict))
d = {
"a": True,
"b": False,
"c": True
}
remap_dict = {}
self.assertDictEqual(d, utils.remap_dict(d, remap_dict))
d = {}
remap_dict = {
"a": "a2",
"b": "b2"
}
self.assertDictEqual(d, utils.remap_dict(d, remap_dict))
d = {
"a": True,
"b": False,
"c": True
}
remap_dict = {
"a": "b",
}
expected = {
"b": False,
"c": True
}
self.assertDictEqual(expected, utils.remap_dict(d, remap_dict))
def predict_minibatch(self, inputs):
"""Predict on a single minibatch of examples."""
model_inputs = [self.preprocess(ex) for ex in inputs]
outputs = self.model.predict_minibatch(model_inputs)
outputs = [utils.remap_dict(mo, self.FIELD_RENAMES) for mo in outputs]
# TODO(gehrmann): temp solution to get ROUGE scores in data table.
for ex, mo in zip(inputs, outputs):
score = self._scorer.score(target=ex["reference"],
prediction=self._get_pred_string(
mo["output_text"]))
mo["rougeL"] = float(score["rougeL"].fmeasure)
return outputs
def predict(self, inputs):
"""Predict on a single minibatch of examples."""
inputs = list(
inputs) # needs to be referenced below, so keep full list
model_inputs = (self.preprocess(ex) for ex in inputs)
outputs = self.wrapped.predict(model_inputs)
outputs = (utils.remap_dict(mo, self.FIELD_RENAMES) for mo in outputs)
# TODO(gehrmann): temp solution to get ROUGE scores in data table.
for ex, mo in zip(inputs, outputs):
score = self._scorer.score(target=ex["reference"],
prediction=self._get_pred_string(
mo["output_text"]))
mo["rougeL"] = float(score["rougeL"].fmeasure)
yield mo
def predict_minibatch(self, inputs):
"""Predict on a single minibatch of examples."""
model_inputs = [self.preprocess(ex) for ex in inputs]
outputs = self.model.predict_minibatch(model_inputs)
return [utils.remap_dict(mo, self.FIELD_RENAMES) for mo in outputs]
def predict(self, inputs):
"""Predict on a single minibatch of examples."""
model_inputs = (self.preprocess(ex) for ex in inputs)
outputs = self.wrapped.predict(model_inputs)
return (utils.remap_dict(mo, self.FIELD_RENAMES) for mo in outputs)