New to ML/NN here. Given a set of integer values over a time series how can I accurately predict the next value in the sequence? Would be supplying the time series data from an MSSQL database and if the presiction algorithm can’t be done via T-SQL then C# would be the fallback langauge I could code. Any auggestions? Even if it just how you’d logically approach it in lieu of a C#/SQL solution.
Sounds like basic Linear Regression to me.
If you're going the deep learning route, look up recurrent neural networks or lstm networks
Lstm is one kind of rnn. Another less known is gru.
Do you know anything specific about the data?
Col1 = date: m/d/yyyy Col2 = whole number: int/bigint I want to handle one of two cases — Case 1: Non-Random data that recurs over time such as min/max temperature or the number of users who visit a webpage on a given day of the year based on historical traffic patterns Case 2: Random sequence where there is a value between a min and max, predicting what number will be next this isn’t necessarily bound to a time series but it is in theory since the random numbers are generated in a sequence and thus can be timestamped. Col1 in this case alternatively an order number. In both cases the output has no decimal Trying to keep these scenarios basic to start for the sake of learning before I apply it to real datasets
Are you trying to predict lotto numbers?
That would involve interdependencies between several columns and with potentially hundreds of millions of combinations in some cases it would be way too complex for what I’m trying to do just looking to learn and understand how this works
Use any sequence modeling algorithm. Traditional ML algorithms like HMM/CRF or Deep Learning algorithms like LSTM/GRU/RNN. Also for traditional time series forecasting, there are various algorithms such as arima etc. You could use any of those to accomplish your task. Good luck!
Benchmark any method that you use against a trellis decoding method
In any ml problem, understand data then extract the most statistically meaningful features to build a model. Blindly throwing data into DNN won’t work well.
Ask Sadhguru
The inner engineering guy?
Yes