Dynet contributors, thank you!

[slawe...@gmail.com]

Hi,

Dynet was the enabler of my winning submission of M4 Forecasting Competition (https://www.m4.unic.ac.cy)

The model is a hybrid one, mixing formulas from an established statistical time series method (Holt-Winters) and an RNN. 

It is also a hierarchical model, as it uses a number of per time series parameters and a global forecasting RNN stack, so a Dynamic Graph NN was needed.

The code and description is available at https://github.com/M4Competition/M4-methods/tree/master/118%20-%20slaweks17. 

BTW, I also coded 3 different dilated LSTMs: standard, with attention, and with special residual connections. They are in the repository I listed above, I will try to submit them to the Dynet, but I need to do some polishing, as my code standard my be questionable :-)

Thank you again guys!

Slawek