XGBoost: Shouldn't data source of model.predict is from past data? about stocks HOT 1 CLOSED

Hi @viper7882,

Thanks for your comments. They are very encouraging and I am happy that you found the projects fun.

I read through your comments carefully, I believe you are talking about this section of the code in the Final model section:

rmse, mape, est = train_pred_eval_model(X_train_cv_scaled, y_train_cv_scaled, X_sample_scaled, y_sample, test['adj_close_mean'], test['adj_close_std'], seed=model_seed, n_estimators=n_estimators_opt, max_depth=max_depth_opt, learning_rate=learning_rate_opt, min_child_weight=min_child_weight_opt, subsample=subsample_opt, colsample_bytree=colsample_bytree_opt, colsample_bylevel=colsample_bylevel_opt, gamma=gamma_opt)

And you said I should replace X_sample_scaled with X_cv_scaled. I don't agree with you this should be the case.

Firstly, please note that the model is intended to predict the value for day t (note we are doing a one-day prediction and not multiple days prediction), based on values from days t-1, t-2, ... and so on. These features of lags t-1, t-2, ... are contained in the X_sample_scaled(= test_scaled[features]). So we are passing the right stuff in.

If you pass in X_cv_scaled, then you are doing model.predict(X_cv_scaled) and obtaining the predictions for X_cv_scaled. But we are not interested in predictions for X_cv_scaled at the 'Final model' section, because we want predictions for X_sample_scaled instead.

Let me know if the above is clear, thanks. Again, thanks for sharing the issue with me.

from stocks.