def predict(self, images: List[Url], simple: bool) -> Prediction:
"""
Predicts using the trained model
"""
filepath = self._names.prediction_cache(images)
mkdirname(filepath)
if os.path.exists(filepath):
print("LOADING: %s" % filepath)
with open(filepath, 'rb') as f:
return dill.load(f)
x = asarray(images)
seq = Sequence1(x, x, self._res, self._batch)
predictions: ndarray = self._kmodel.predict_generator(generator=seq,
verbose=1)
if simple:
if predictions.ndim == 2 and predictions.shape[1] == 1:
predictions = predictions.flatten()
y = predictions.tolist()
if simple:
y = self._data.translate_predictions(y)
results = Prediction(images, y)
print("SAVING: %s" % filepath)
with open(filepath, 'wb') as f:
dill.dump(results, f)
return results
def run_cached(self, images: List[Url], **kwargs) -> ClusterResults:
filepath = self._cache_path(images, **kwargs)
mkdirname(filepath)
if os.path.exists(filepath):
print('LOADING: %s' % filepath)
with open(filepath, 'rb') as f:
return dill.load(f)
results = self.run(images, **kwargs)
print('SAVING: %s' % filepath)
with open(filepath, 'wb') as f:
dill.dump(results, f)
return results
def evaluate_test_set(self) -> Dict[str, float]:
"""
Evaluates the model using the test set
"""
filepath = self._names.test_set_evaluation_cache()
if os.path.exists(filepath):
print("LOADING: %s" % filepath)
with open(filepath, 'rb') as f:
return dill.load(f)
x = self._data.test().x().load()
y = self._data.test().y().load()
results = self.evaluate(x, y)
mkdirname(filepath)
print("SAVING: %s" % filepath)
with open(filepath, 'wb') as f:
dill.dump(results, f)
return results