def return_summary(request: Request,
input_text: str = Form(...),
summary_len: int = Form(150),
beams: int = Form(2)):
summary = generate_summary(input_text, summary_len, beams, model,
tokenizer)
return templates.TemplateResponse('index.html',
context={
'request': request,
'summary': summary,
'text': input_text
})
def get_summary():
""" Returns a summary of an article found at a given url """
# Parse url from form response
req = request.form.to_dict()
memory = json.loads(req['Memory'])
url = memory['twilio']['collected_data']['summarize_article']['answers']['article_url']['answer']
# Parse article text
text = parse_article(url)
# Generate summary
summary = generate_summary(text)
# Return summary
actions = {
"actions": [
{
"say": summary
}
]
}
return actions
with h5py.File(PATH_TVSum, 'r') as hdf:
for video_name in keys:
video_index = video_name[6:]
user_summary = np.array(
hdf.get('video_' + video_index + '/user_summary'))
sb = np.array(hdf.get('video_' + video_index + '/change_points'))
n_frames = np.array(hdf.get('video_' + video_index + '/n_frames'))
positions = np.array(hdf.get('video_' + video_index + '/picks'))
all_user_summary.append(user_summary)
all_shot_bound.append(sb)
all_nframes.append(n_frames)
all_positions.append(positions)
all_summaries = generate_summary(all_shot_bound, all_scores, all_nframes,
all_positions)
all_f_scores = []
# compare the resulting summary with the ground truth one, for each video
for video_index in range(len(all_summaries)):
summary = all_summaries[video_index]
user_summary = all_user_summary[video_index]
f_score = evaluate_summary(summary, user_summary, eval_method)
all_f_scores.append(f_score)
f_score_epochs.append(np.mean(all_f_scores))
print("f_score: ", np.mean(all_f_scores))
with open(path + '/f_scores.txt', 'w') as outfile:
json.dump(f_score_epochs, outfile)
import generate_summary as gs
if __name__ == "__main__":
top_n = int(input("Enter the number of sentences you want in the summary:"))
gs.generate_summary( "input.txt",top_n)
else:
if args.model_path == "":
raise FileNotFoundError
model = TransformerSummarizer(ATTENTION_HEADS, N_LAYERS, N_LAYERS, DIM_FEEDFORWARD, \
SEQ_LEN, VOCAB_SIZE, PAD_IDX, src_list, d_model=D_MODEL).to(device)
model.load_state_dict(torch.load(args.model_path, map_location=device))
Path.mkdir(out_dir, parents=True, exist_ok=True)
with open(os.path.join(out_dir, "raw.txt"), "w", encoding="utf-8") as text, \
open(os.path.join(out_dir, "pred.txt"), "w", encoding="utf-8") as pred, \
open(os.path.join(out_dir, "true.txt"), "w", encoding="utf-8") as true:
for data in tqdm(test_data, total=TEST_SIZE):
src_text = data.text
trg_text = data.summ
raw_text = " ".join(src_text)
true_summary = " ".join(trg_text)
prediction = generate_summary(raw_text, model, src_list, src_dict,
MAX_LENGTH)
text.write(raw_text + "\n")
text.flush()
true.write(true_summary + "\n")
true.flush()
pred.write(prediction + "\n")
pred.flush()