right now, all price changes are represented relatively to the previous price. i should look into scaling them for each prediction - that is, get data for the current sequence, choose min and max price and scale the target between them. the logic is that if you are looking at a time period and predicting from it, it makes no sense to care about the relevance of this number to the nominal price, but rather to the size of the change in the space of the current "market tunnel", if you will.

• measure accuracy!

thank you chatgpt for this wonderful idea. quoting:

Voting Classifier/Regressor
In a voting classifier or regressor, each base model predicts the target outcome, and the final prediction is determined by a majority vote (for classification) or an average (for regression) of the individual predictions.
Bagging (Bootstrap Aggregating)
Bagging involves training multiple instances of the same base model on different subsets of the training data (bootstrap samples) and then aggregating their predictions (e.g., averaging for regression, voting for classification).
Random Forest
A random forest is an extension of bagging that specifically applies to decision trees. It constructs an ensemble of decision trees, each trained on a different bootstrap sample and a random subset of features.
Boosting
Boosting focuses on sequentially training weak base models, where each subsequent model corrects the errors of the previous one. Common boosting algorithms include AdaBoost, Gradient Boosting, and XGBoost.
Stacking
Stacking involves training multiple diverse base models and using a meta-model (often another model) to learn how to combine their predictions optimally. The meta-model is trained on the predictions of the base models.
AdaBoost (Adaptive Boosting)
AdaBoost assigns weights to the training instances and emphasizes the misclassified instances to allow subsequent base models to focus more on those cases and improve overall prediction.
Gradient Boosting Machines (GBM)
GBM builds an ensemble of decision trees sequentially, where each tree corrects the errors of the previous ones. It fits each tree to the residuals of the previous tree.
LightGBM and XGBoost
These are popular implementations of gradient boosting, optimized for speed and performance. They use advanced techniques to build accurate ensembles of decision trees.

https://keras.io/api/layers/attention_layers/attention/
https://www.youtube.com/watch?v=MNSmOih_pmg

research techniques:

• dropout
• proper validation
• model regularization