Lesson 4: Structured Time Series Implementation question


(Khunakorn Luyaphan) #1

Dear all

I have a question about the lesson 4 if anyone could help explain
I am trying to implement the concept to practice on a kaggle competition,


which I suppose is also time series

In the lesson, we are trying to predict multiple target into the future (2 weeks)
What should I do if I only have a single target for each label, something like

Training Set (Let say the target is the avg sales of that store / how well the store is doing)

Store 1 Day 1 Sales 10000,
Store 1 Day 2 Sales 7000,
Store 1 Day 3 Sales 5000,               Label 1 Training Target = 7333

Store 2 Day 1 Sales 1000,
Store 2 Day 2 Sales 2000,
Store 2 Day 3 Sales 1500,
Store 2 Day 4 Sales 2050,               Label 2 Training Target = 1637

Test Set ( Trying to output a single feature from all of the time series of Store 3 )

Store 3 Day 1 Sales 1000,
Store 3 Day 2 Sales 2000, 
Store 3 Day 3 Sales 1500,               Label 3 Test Target = ????   

Right now What I’m doing is copying the target into every row in the training set

Training Set

Store 1 Day 1 Sales 10000,              Label 1 Training Target = 7333
Store 1 Day 2 Sales 7000,               Label 1 Training Target = 7333
Store 1 Day 3 Sales 5000,               Label 1 Training Target = 7333

Store 2 Day 1 Sales 1000,               Label 2 Training Target = 1637
Store 2 Day 2 Sales 2000,               Label 2 Training Target = 1637
Store 2 Day 3 Sales 1500,               Label 2 Training Target = 1637
Store 2 Day 4 Sales 2050,               Label 2 Training Target = 1637

And even on training the results is pretty bad