I’m trying to implement RNN in pytorch but there are some topic i just dont understand:

Suppose I have a time series input of 150 days, each day has 10 features.I want LSTM to predict next day value of one features when i give it one day input. How can i format it and put it in RNN’s foward part?

I’ve known that LSTM has hidden_size as a argument. What does it do? As far as i know
hidden_state = tanh(cell)*sigmoid(weighted input, prev_hidden)
which is just a number and have size of 1. 
Because I want it to predict one feature’s value of the next day. How could i define loss for it?