Label-mixing transforms

Callbacks that perform data augmentation by mixing samples in different ways.

---

source

MixHandler1d

 MixHandler1d (alpha=0.5)

A handler class for implementing mixed sample data augmentation

---

source

MixUp1d

 MixUp1d (alpha=0.4)

Implementation of https://arxiv.org/abs/1710.09412

from fastai.learner import *
from tsai.models.InceptionTime import *
from tsai.data.external import get_UCR_data
from tsai.data.core import get_ts_dls, TSCategorize
from tsai.data.preprocessing import TSStandardize
from tsai.learner import ts_learner

X, y, splits = get_UCR_data('NATOPS', return_split=False)
tfms = [None, TSCategorize()]
batch_tfms = TSStandardize()
dls = get_ts_dls(X, y, tfms=tfms, splits=splits, batch_tfms=batch_tfms)
learn = ts_learner(dls, InceptionTime, cbs=MixUp1d(0.4))
learn.fit_one_cycle(1)

epoch	train_loss	valid_loss	time
0	1.908455	1.811908	00:03

---

source

CutMix1d

 CutMix1d (alpha=1.0)

Implementation of https://arxiv.org/abs/1905.04899

---

source

IntraClassCutMix1d

 IntraClassCutMix1d (alpha=1.0)

Implementation of CutMix applied to examples of the same class

X, y, splits = get_UCR_data('NATOPS', split_data=False)
tfms = [None, TSCategorize()]
batch_tfms = TSStandardize()
dls = get_ts_dls(X, y, tfms=tfms, splits=splits, batch_tfms=batch_tfms)
learn = ts_learner(dls, InceptionTime, cbs=IntraClassCutMix1d())
learn.fit_one_cycle(1)

epoch	train_loss	valid_loss	time
0	1.813483	1.792010	00:03

X, y, splits = get_UCR_data('NATOPS', split_data=False)
tfms = [None, TSCategorize()]
batch_tfms = TSStandardize()
dls = get_ts_dls(X, y, tfms=tfms, splits=splits, batch_tfms=batch_tfms)
learn = ts_learner(dls, cbs=CutMix1d(1.))
learn.fit_one_cycle(1)

epoch	train_loss	valid_loss	time
0	1.824509	1.774964	00:04