the most important points? 1. CV strategy (vote >= 10) 2. Using various representations: raw EEG(1d, 2d), Spectrogram(STFT, CWT, Superlet) 3. Replacing Softmax with Entmax •How did you select features? • We checked CV score for selecting input representations and 2d backbones. •Did you make any important feature transformations? • Frequency transformation • Treating raw EEG as the 2D image •Did you use external data? (if permitted) • No, we didn't use any external data.
(vote >= 10)~ GitHub - bdsp-core/IIIC-SPaRCNet Since the test data has more than 10 votes, it is necessary to adjust the validation data accordingly. # Most other participants also followed this assumption.
with Entmax~ • Assumption • The output of softmax is not zero for every target variable, but many training data labels have zero probability classes -> Sparse output may improve the score • What we did • Replacing softmax with entmax only during inference • entmax is the function that is like softmax, but entmax can generate more sparse output • We use α = 1.03 • Improving both public and private LB score ~0.004 [-2, 0, 0.5] softmax [0.049, 0.359, 0.592] entmax (α=1.5) [0.0, 0.326, 0.674] entmax (α=2.0) [0.0, 0.25, 0.75 ] = entmax (α=1.0)