deep-news-summarization
deep-news-summarization copied to clipboard
Perplexity
I have been training the model with a data set that has been provided for a couple of days. Its perplexity is kind of reducing or constant somewhere near to 1.27-1.28. What should be the ideal value to get better results?
Hi, your model perplexity is already very good. A perplexity of 1.27 means that your model have learn a good language model such that there are in average 1/1.27 probability to predict the correct next word in your output sentence.
0
security
security passports orchestras Poisoned `security orchestras orchestras passports orchestras
Canada's TSX Group pursues stronger ties with Chinese stock markets
TSX Group Inc. hopes to pursue stronger ties with the Chinese stock market in a bid to encourage more Chinese companies to list on Canada's dominant stock exchange.
0 Bulldozer Bulldozer NATASHA Bulldozer NATASHA Bulldozer Yardstick Yardstick Pennies toppled Pennies muster unsuccessfully Pennies Pennies Lakes Pennies Pennies Depresses Pennies
A calamitous year sees European police forge closer ties to fight terrorists, pedophiles and hooligans
"Jolted by the carnage of the Madrid train bombings and the horrific cross-border rape and killing spree of a French pedophile, EU states are coming together on an ambitious new vision for policing Europe. It involves joint investigation teams and the sharing of hitherto jealously guarded criminal records and crime-fighting technology."
----------------------------------This is kind of evaluation I am getting. Everything seems a bit scrambled.