Hello all 
nice to meet you
I need to develop a tool that, given a customer id, predicts how many truck visits they will receive in the next hour.
I have several years worth of historical data for customers, so I think that a Deep Learning mechanism might be well suited for the task and I’d like to use PyTorch to build it. I have only taken the Udacity course on PyTorch so never attempted a real-world application with it and would really really appreciate some guidance.
I think that for now, I won’t be using many data points. Possibly only customer id and historical truck visits for every hour of the day. I can’t really think of how else I’d make such a prediction (weather maybe? But I don’t have the weather data at hand so probably best to skip for the time being)
I have several questions in this regard:
-
Am I right to assume an LSTM is best suited for this time-based prediction scenario?
-
If I keep getting new truck visit data, how can I include the new data into the prediction engine? Do I need to retrain every so often or are there techniques to include the incoming data in the next prediction as it comes?
-
What would you use as a general strategy to solve this problem? Are there some tutorials out there that you could link me to that try to do something similar?
Again, any help is much appreciated.
Thanks in advance! 