Time-Series-Transformers

M4-Competition Data

git clone https://github.com/Mcompetitions/M4-methods.git

Backlog

• Implement M4-Competition inference procedure with a Baseline Model

References

• https://medium.com/mlearning-ai/transformer-implementation-for-time-series-forecasting-a9db2db5c820

Starting train. 272 batches 216888/800 Epoch 1/100 [497.712secs] -> Train loss: 0.04742 Epoch 2/100 [501.675secs] -> Train loss: 0.01139 Epoch 3/100 [506.052secs] -> Train loss: 0.00809 Epoch 4/100 [506.472secs] -> Train loss: 0.00696 Epoch 5/100 [506.294secs] -> Train loss: 0.00636 Epoch 6/100 [505.678secs] -> Train loss: 0.00599 Epoch 7/100 [504.128secs] -> Train loss: 0.00576

checkpoint

Epoch 8/100 [455.912secs] -> Train loss: 0.00629 Epoch 9/100 [458.781secs] -> Train loss: 0.00549 Epoch 10/100 [451.579secs] -> Train loss: 0.00537 Epoch 11/100 [451.383secs] -> Train loss: 0.00528 Epoch 12/100 [451.732secs] -> Train loss: 0.00521 Epoch 13/100 [451.359secs] -> Train loss: 0.00515 Epoch 14/100 [451.380secs] -> Train loss: 0.00510 Epoch 15/100 [451.347secs] -> Train loss: 0.00506 Epoch 16/100 [451.356secs] -> Train loss: 0.00502 Epoch 17/100 [451.356secs] -> Train loss: 0.00499 Epoch 18/100 [451.353secs] -> Train loss: 0.00495 Epoch 19/100 [451.377secs] -> Train loss: 0.00494 Epoch 20/100 [451.343secs] -> Train loss: 0.00491 Epoch 21/100 [451.353secs] -> Train loss: 0.00489