Hi dimitry,
I really appreciate this article series very much! Thank you for that! But please help with this problem:
If I load the file "EURUSD_PERIOD_H1_rnn_vae.nn" as mentioned in this article, I get the message "Error of load model" and "The file is damaged":
If I trace, I find the loading fail in this line in NeuroNet.mqh:
If I load the model from part 23, named "EURUSD_i_PERIOD_H1_test_rnn.nnw" it seems to work, but this model only has two layers. This is not the right one. Did I miss something??
If I load the file "EURUSD_PERIOD_H1_rnn_vae.nn" as mentioned in this article, I get the message "Error of load model" and "The file is damaged":
If I load the model from part 23, named "EURUSD_i_PERIOD_H1_test_rnn.nnw" it seems to work, but it only has two layers. This is not the right one. Did I miss something??
Hi,
For load file "EURUSD_PERIOD_H1_rnn_vae.nnw" you need recompilation NetCreator with new NeuroNet.mqh library. In last model we replace CBufferDouble to CBufferFloat. And add some types of layer.
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
New article Neural networks made easy (Part 25): Practicing Transfer Learning has been published:
In the last two articles, we developed a tool for creating and editing neural network models. Now it is time to evaluate the potential use of Transfer Learning technology using practical examples.
Testing results are shown in the chart below. As you can see, the pre-trained model started with a smaller error. But soon the second model caught up and further their values were quite close. This confirms the earlier conclusion that the encoder architecture has a significant impact on the performance of the entire model.
Pay attention to the learning rates. The pre-trained model required six times less time to pass one epoch. Of course, this is the pure time, without taking into account the autoencoder training.
Author: Dmitriy Gizlyk