ANN Shape for Sale Prediction

The inputs (features) and expected output for my ANN are these:

• Input 1: Product id (number, cast to double)
• Input 2: Year in the past (1900..2017, cast to double)
• Input 3: Month of year (1..12, cast to double)
• Expected output: Sale of month (number of units sold, cast to double)

I need to predict the sale of a product for a certain month in a certain year. How many layers and how many neurons on there layers should I put?

There's one suggestion in here: https://ai.stackexchange.com/questions/3292/ann-shape-for-sale-prediction

But I finally implemented the network as LMS (least mean square) neural network with:

• 4 inputs
• 9 neurons in hidden layer
• 1 neuron in output layer

And instead of using 3 feature inputs, I use the sale of 4 previous months of a product as input. The prediction curve is rather flat than it's supposed to be but at least it is working.