jinglescode/time-series-forecasting-pytorch

is this model any good?

ren85 opened this issue · 3 comments

ren85 commented

If I were to build a model that predicts yesterdays closing price as today's closing price we would also get similar graph, but the model is useless. Where are prediction accuracy numbers? I bet they would be terrible.

Hi @ren85,
As we are prediction values, we used MSE. You can check the MSE from the logs, for example, on the training and validation set, where loss is the MSE:

Epoch[100/100] | loss train:0.006102, test:0.000972 | lr:0.000100
ren85 commented

The question is is this any better than using yesterdays closing price as today's prediction

is this any better than using yesterdays closing price as today's prediction

To find that out, what you can do is to calculate the error using MSE of yesterdays closing price and today's prediction, and then compare it against the MSE of the model after training.