trainX_text = trainX_text.reset_index(drop=True)
testX_text = testX_text.reset_index(drop=True)
#load text model and predict
train_data = trainX_text
test_data = testX_text
text_results_train = []
for i in range(len(train_data)):
result = predict_text(bert_model,train_data[i])
text_results_train.append(result)
text_results_test = []
for i in range(len(test_data)):
result = predict_text(bert_model,test_data[i])
text_results_test.append(result)
# text results
# unlist
text_results_train_cat = np.concatenate(text_results_train, axis=0)
text_results_test_cat = np.concatenate(text_results_test, axis=0 )
text_results_train_cat = split(text_results_train_cat, 2.86,3.59)
text_results_test_cat = split(text_results_test_cat, 2.86,3.59)
rev = { v:k for k,v in index_map.items()}
##### audio
# convert audio into feature
audio = audio2wave(file)
audio_X = prepare_data_librosa(audio,
features='logmel',
scaled=True)
stackedX_test = stacked_dataset(all_models, audio_X)
audio_pred = audio_logistic.predict(stackedX_test)
print('Audio prediction:', [rev[item] for item in audio_pred])
##### text
text_pred = predict_text(bert_model,text)
print('Text prediction:', [rev[item] for item in np.array([np.argmax(text_pred)])])
##### ensemble
ensemble_text_test = softmax(text_pred)
ensemble_audio_test = audio_logistic.predict_proba(stackedX_test)
stack_test = np.dstack((ensemble_text_test, ensemble_audio_test))
stack_test = stack_test.reshape((stack_test.shape[0], stack_test.shape[1]*stack_test.shape[2]))
ensemble_pred = final_logistic.predict(stack_test)
print('Ensemble prediction:', [rev[item] for item in ensemble_pred])
def cate(emo):
if emo == 'hap':
i = 'pos'