Time series structure data with LSTM

[kevin_d]

Hi everyone,

I learnt how to use DL method to deal with structure data in the course, and want to use the method to deal with this (challenge)[https://www.kaggle.com/c/recruit-restaurant-visitor-forecasting]. I just found that the kernels on LB would like to leverage all kinds statistics across different time frames to predict visitors, and am wondering is that appropriate to combine fast.ai structure data processing pipeline and LSTM in order to preventing tricky feature engineering ? Thanks.

[Brad_S]

of interest?

Sales (momentum) for future prediction? (Lesson3-4 Rossman)

Sorry I saw that in the other thread but wasn’t sure what you were saying. Upon re-reading, I think I understand. What you’re referring to is known as “auto-regression”. That is, using earlier time periods of the dependent variable as predictors. It is indeed common in time series, but not directly doable with the kind of model used in this notebook. For that, you’d need an RNN, which is introduced next! (However, it seems that in practice RNNs aren’t giving as good results as fu…

[kevin_d]

Thank you. It looks like we have similiar questions. These days I tried out some kernels on leaderboard, and guess that even thought LSTM is able to memorise previous status of a system, it is not able to figure out appropriate feature representation for a centain problem if there are not enough data. Thus, I am still unable to find a ’ silver bullet ’ for time series structure data problems.