Visualize on Tableau to detect data patterns and correlations. Produce 128 features based on original features.
QYQ: Applied naive Gaussian Process Regression with Matern, Rational Quadratic, White Kernel, memory explosion Extract non-linear feature map with kernel approximation (reduction from 45 to 40 features), applied Bayesian Ridge Regression, validation score = 5.96 Directly apply Hist Gradient Boosting Regressor, validation score = 5.89
LB & LKL Built a fully connected neural network with 2 layers and train, validation score = 5.96
split_day = 435
df_train = df[df["date_id"] <= split_day]
df_valid = df[df["date_id"] > split_day]
print(f"train : {df_train.shape}, valid : {df_valid.shape}")
Add feature selection.