How do we have to go about this problem?
I tried making a 2D array, with every bracket of consecutive 500 entries in the training data, padding using keras.preprocessing.sequence, and training.
The layers are embedding, RNN, Dense(sigmoid).
When I try to train, the loss goes negative and accuracy remains 0.
hey @Aayushhh ,
it depends upon your thinking in what way you wanna solve this problem.
It can also converted into tabular data and use simple regression algorithms to do so.
Or the way you implemented it. But its a regression task , not a classification task. So your accuracy will be very very low ,may be 0 also. The way you are implementing is correct , but you will need to work more on it.
Like its not necessary to take previous 500 days record , sometime last month or last week also works.
output layer need to be linear ,not sigmoid.
sigmoid return probabilities.
So just search about it a bit , understand the task and then implement it. It will help you .
I hope this helped.
Thank You 
.