def get_x_data_features_by_sampleidlist(self, cur_sampid_list):
"""
Do features extraction and cached, get x_data array by cur_sampid_list.
:param cur_sampid_list:
:return: list of arrays.
"""
if self.spec_len_status == 1:
self.g_cached_train_x.clear()
self.g_cached_x_sampldid_list.clear()
self.spec_len_status = 2
cur_need_id_list = [
sampleid for sampleid in cur_sampid_list
if sampleid not in self.g_cached_x_sampldid_list
]
if len(cur_need_id_list) > 0:
cur_need_origin_xdata = [
self.g_ori_train_x[cur_sampid]
for cur_sampid in cur_need_id_list
]
self.imp_feat_args["mode"] = "train"
cur_need_x_data_list = ut.pre_trans_wav_update(
cur_need_origin_xdata, self.imp_feat_args)
for i, cur_sampid in enumerate(cur_need_id_list):
self.g_cached_x_sampldid_list.extend(cur_sampid_list)
self.g_cached_x_sampldid_list = list(
set(self.g_cached_x_sampldid_list))
self.g_cached_train_x[cur_sampid] = cur_need_x_data_list[i]
return [
self.g_cached_train_x[cur_sampid] for cur_sampid in cur_sampid_list
]
def test(self, test_x, remaining_time_budget=None):
"""
:param test_x: list of vectors, input test speech raw data.
:param remaining_time_budget:
:return: A `numpy.ndarray` matrix of shape (sample_count, class_num).
here `sample_count` is the number of examples in this dataset as train
set and `class_num` is the same as the class_num in metadata. The
values should be binary.
"""
K.set_learning_phase(0)
model = self.model
if self.test_idx == 1 or self.spec_len_status == 2:
self.test_x_data_report = self.autospeech_eda.get_x_data_report(
x_data=test_x)
or_data_x_seq_mean = self.test_x_data_report.get("x_seq_len_mean")
self.spec_len_status = 0
self.imp_feat_args["mode"] = "test"
test_x = ut.pre_trans_wav_update(test_x, self.imp_feat_args)
test_x = np.array(test_x)
self.test_x = test_x[:, :, :, np.newaxis]
if self.good_to_predict():
y_pred = model.predict(self.test_x,
batch_size=self.tr34_mconfig.PRED_SIZE)
if self.tr34_mconfig.ENABLE_PRE_ENSE and self.decision_if_pre_ensemble(
):
self.best_test_predict_prob_list[self.round_idx - 1] = y_pred
y_pred = self.test_pred_ensemble()
self.test_idx += 1
self.last_y_pred = y_pred
self.last_y_pred_round = self.round_idx
return y_pred
else:
return self.last_y_pred
def decide_if_full_valid(self):
if self.round_idx == self.tr34_mconfig.FULL_VAL_R_START:
self.accpet_loss_list = [100]
self.imp_feat_args["mode"] = "train"
self.g_valid_x_features = ut.pre_trans_wav_update(self.g_valid_x, self.imp_feat_args)
self.g_valid_gen = generator.DataGenerator(self.g_valid_x_features, self.g_valid_y, **self.params)
if self.round_idx > self.tr34_mconfig.FULL_VAL_R_START:
return True
else:
return False