Discussing the article: "Neural networks made easy (Part 63): Unsupervised Pretraining for Decision Transformer (PDT)"
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: Neural networks made easy (Part 63): Unsupervised Pretraining for Decision Transformer (PDT).
For the fine-tuning period, I needed several dozen successive iterations of downstream training and testing, which also required time and effort.
However, the learning results were not so promising. As a result of training, I received a model that trades a minimum lot with varying success. In some parts of history, the balance line showed a clear upward trend. In others it was a clear decline. In general, the model results both on training data and on a new set were close to 0.
Positive aspects include the model's ability to transfer the experience gained to new data, which is confirmed by the comparability of testing results on the historical dataset of the training set and on the following history interval. In addition, you can see that the size of a profitable trade is considerably greater than that of a losing one. In both historical data segments, we observe that the size of the average winning trade exceeds the maximum loss. However, all the positive aspects are offset by the low share of profitable trades, which is just under 40% in both historical intervals.
For the fine-tuning period, I needed several dozen successive iterations of downstream training and testing, which also required time and effort.
However, the learning results were not so promising. As a result of training, I received a model that trades a minimum lot with varying success. In some parts of history, the balance line showed a clear upward trend. In others it was a clear decline. In general, the model results both on training data and on a new set were close to 0.
Positive aspects include the model's ability to transfer the experience gained to new data, which is confirmed by the comparability of testing results on the historical dataset of the training set and on the following history interval. In addition, you can see that the size of a profitable trade is considerably greater than that of a losing one. In both historical data segments, we observe that the size of the average winning trade exceeds the maximum loss. However, all the positive aspects are offset by the low share of profitable trades, which is just under 40% in both historical intervals.
Author: Dmitriy Gizlyk