Recurrrent neural network - RNN
LTA/STA - Long Term Averaging och Short Term Averaging - flyttad till features
~~Behövs?? Kanske det, viktningsfunktion/convulution? ~~ Avfärdad då tvälingsdata är osorterade 150k bitar.
Plocka ut 150K bitar på random för att validera modellen så man ser var den gör bra ifrån sig och var den gör dåligt
Preprocessing med STA/LTA kommer att behövas. - Avfärdad
Vad händer om vi tränar modellen på random utvalda 150k bitar?
mean
median(abs)
std
max
min
Classic LTA/STA
Recursive LTA/STA??
View tensorboard with: tensorboard --logdir=Graph --host=localhost --port=8088
| model | features | dropout | epochs train | activation fun | train mae | test mae |
|---|---|---|---|---|---|---|
| simpleffnn | statistical features1 | 0.25 | 1000 | tanh | 1.2000000000000 | 1.800 |
| simpleffnn | statistical features1 | 0.25 | 100 | tanh | 1.7000000000000 | 1.672 |
| simpleffnn | statistical features1 | 0.25 | 10 | tanh | 2.0029595604348 | 1.537 |
| simpleffnn | statistical features1 | 0.25 | 3 | tanh | 2.0029595604348 | 1.540 |
| simpleffnn | statistical features1 | 0.5 | 1000 | tanh | 1.6404929039735 | 1.650 |
| simpleffnn | statistical features1 | 0.5 | 100 | tanh | 1.9273089859860 | 1.596 |
| simpleffnn | statistical features1 | 0.5 | 10 | tanh | 2.0137530461991 | 1.531 |
| simpleffnn | statistical features1 | 0.5 | 3 | tanh | 2.0259630969190 | 1.497 |
| -------------- | ----------------------- | --------------- | --------------- | --------------- | --------------- | ---------- |
| simpleffnn | statistical features3 | 0.25 | 1000 | tanh | 1.3058205365700 | 1.830 |
| simpleffnn | statistical features3 | 0.25 | 100 | tanh | 1.8733245993213 | 1.622 |
| simpleffnn | statistical features3 | 0.25 | 10 | tanh | 2.0358140996289 | 1.522 |
| simpleffnn | statistical features3 | 0.25 | 3 | tanh | 2.0310845783116 | 1.489 |
| simpleffnn | statistical features3 | 0.5 | 10 | tanh | 2.0118357298538 | 1.513 |
| simpleffnn | statistical features3 | 0.5 | 3 | tanh | 2.0409161202138 | 1.471 |
| simpleffnn | statistical features3 | 0.6 | 3 | tanh | 2.0442734291910 | 1.483 |
| -------------- | ----------------------- | --------------- | --------------- | --------------- | --------------- | ---------- |
| simpleffnn~10 | statistical features3 | 0.25 | 3 | tanh | 2.0229714746295 | 1.484 |
| simpleffnn~100 | statistical features3 | 0.5 | 3 | tanh | 2.0308562487956 | 1.469 |
| simpleffnn~10 | statistical features3 | 0.5 | 3 | tanh | 2.0319643617985 | 1.473 |
| -------------- | ----------------------- | --------------- | --------------- | --------------- | --------------- | ---------- |
| heavy_ffn | statistical features3 | 0.5 | 10 | relu | 1.9859311313708 | 1.516 |
| heavy_ffn | statistical features3 | 0.5 | 3 | relu | 2.0442280633093 | 1.600 |
| -------------- | ----------------------- | --------------- | --------------- | --------------- | --------------- | ---------- |
| simpleffnn | statistical features4 | 0.5 | 10 | tanh | 2.0113052635706 | 1.489 |
| simpleffnn | statistical features4 | 0.5 | 3 | tanh | 2.0373891132104 | 1.475 |
| simpleffnn | statistical features5 | 0.5 | 10 | tanh | 2.0246831776283 | 1.510 |
| simpleffnn | statistical features4 | 0.5 | 3 | tanh | 2.0402858391355 | 1.491 |
| name | performance |
|---|---|
| submission |
1.451 |