# 152th solution

Overview
First, I would like to express my sincere thanks to the organizers of the Child Mind Institute - Detect Sleep States Competition.
I tried this competition based on
[@tubotubo](https://www.kaggle.com/tubotubo)
's baseline. I extend my gratitude.
My Solution
Changed the way of validation
Divided into groups using GroupKFold, ensuring an equal distribution of number of events.
0.727
Clipped enmo
Clipped at 1 as outliers were significant
0.724
Seed averaging with three different values
0: 0.725
21: 0.724
42: 0.723
Ensamble models
Averaged the predictions of each model, and then applied post-processing using find_peaks.
0.73~0.735
Did Not Work
window for input was expanded from 8 hours to 24 hours
I wanted to include information from the previous day and other times in the models.
Changing parameters of Gaussian label
Attempted to pass smoother labels to the model, but did not see an improvement in accuracy.
Changing parameters of find_peaks
Deleting low score predictions in post-process
Set a threshold for scores, but it was not successful.