# Do we have anything like accuracy and loss in RNN models?

I have a paper about trading which has been implemented with RNN on Tensorflow. We have about 2 years of data from trading. Here are some samples :

Date, Open, High, Low, Last, Close, Total Trade Quantity, Turnover (Lacs)

2004-08-25 , 1198.7, 1198.7, 979.0, 985.0, 987.95, 17116372.0, 172587.61

2004-08-26 , 992.0, 997.0, 975.3, 976.85, 979.0, 5055400.0, 49828.65

I need to predict the the future of trading (for example, the latest 10 days ). So, how can I make sure that my model is working correctly. Do we have any "accuracy" or "loss" like what we have in Deep Learning?

## 1 Answer

RNN's stand for Recurrent Neural Networks which is, in fact, Deep Learning.

There has to be a loss since you're dealing with supervised learning and the typical loss metrics used are the same as you would see in feedforward networks (usually binary cross-entropy), the main difference being loss would be calculated between the true label at a particular time stamp $$(t)$$ and the prediction made from the subset of the network until time-stamp $$(t-1)$$. This leads the loss to act on all timestamps.

Accuracy metrics also would be used in the same way such as Mean Square Error or L1. For more details you can go through this link.

Hope this was helpful!

• Yes, it was really helpful and I have another question relating to RL. If I want to reach the maximum total return (profit), according to what we have in Reinforcement Learning and regarding this kind of data set (Stock or Trading), can we do that? If yes, how can we make sure that the return value is the best total maximum return . – Mahdi Amrollahi Nov 25 '19 at 16:08
• Sorry to say I can't say much when it comes to RL. I think it'd be best if you were to ask this in another question. It would make it visible to people who can answer this satisfactorily – ashenoy Nov 25 '19 at 16:54