ROCKET Pytorch

ROCKET (RandOm Convolutional KErnel Transform) functions for univariate and multivariate time series developed in Pytorch.

---

source

ROCKET

 ROCKET (c_in, seq_len, n_kernels=10000, kss=[7, 9, 11], device=None,
         verbose=False)

RandOm Convolutional KErnel Transform

ROCKET is a GPU Pytorch implementation of the ROCKET functions generate_kernels and apply_kernels that can be used with univariate and multivariate time series.

---

source

create_rocket_features

 create_rocket_features (dl, model, verbose=False)

Args: model : ROCKET model instance dl : single TSDataLoader (for example dls.train or dls.valid)

bs = 16
c_in = 7  # aka channels, features, variables, dimensions
c_out = 2
seq_len = 15
xb = torch.randn(bs, c_in, seq_len).to(default_device())

m = ROCKET(c_in, seq_len, n_kernels=1_000, kss=[7, 9, 11]) # 1_000 for testing with a cpu. Default is 10k with a gpu!
test_eq(m(xb).shape, [bs, 2_000])

from tsai.data.all import *
from tsai.models.utils import *

X, y, splits = get_UCR_data('OliveOil', split_data=False)
tfms = [None, TSRegression()]
batch_tfms = TSStandardize(by_var=True)
dls = get_ts_dls(X, y, splits=splits, tfms=tfms, batch_tfms=batch_tfms, shuffle_train=False, drop_last=False)
model = build_ts_model(ROCKET, dls=dls, n_kernels=1_000) # 1_000 for testing with a cpu. Default is 10k with a gpu!
X_train, y_train = create_rocket_features(dls.train, model) 
X_valid, y_valid = create_rocket_features(dls.valid, model)
X_train.shape, X_valid.shape

((30, 2000), (30, 2000))