Discussion of article "Neural networks made easy (Part 21): Variational autoencoders (VAE)"
Hello,
when I compile the NeuroNet.mqh file attached at the end of this article, I get 6 errors, all of them reporting: 'pow' - ambiguous call to overloaded function. Particular lines are 3848, 4468, 6868. Can somebody help me with that please?
Thank you very much
jirivokurka #:
Try adding float before t Hello,
when I compile the NeuroNet.mqh file attached at the end of this article, I get 6 errors, all of them reporting: 'pow' - ambiguous call to overloaded function. Particular lines are 3848, 4468, 6868. Can somebody help me with that please?
Thank you very much
lt = (float)(eta * sqrt(1 - pow(b2, (float)t)) / (1 - pow(b1, (float)t)));
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
New article Neural networks made easy (Part 21): Variational autoencoders (VAE) has been published:
In the last article, we got acquainted with the Autoencoder algorithm. Like any other algorithm, it has its advantages and disadvantages. In its original implementation, the autoenctoder is used to separate the objects from the training sample as much as possible. This time we will talk about how to deal with some of its disadvantages.
To test the operation of the variational autoencoder, we will use the model from the previous articles. Saved it in a new file "vae.mq5". In that model, the encoder returned 2 values on the 5th neural layer. To properly organize the operation of the variational autoencoder, I increased the layer size at the encoder output to 4 neurons. I also inserted our new neural layer working with the latent state of the variational autoencoder as the 6th neuron. The model was trained on EURUSD data and the H1 timeframe without changing the parameters. The last 15 years were used as the time period for model training. A comparative graph of the learning dynamics of multilayer and variational autoencoders is shown in the figure below.
As you can see, according to the results of model training, the variational autoencoder showed a significantly lower data recovery error throughout the entire training period. In addition, the variational autoencoder showed a higher error reduction dynamics.
Based on the test results, we can conclude that for solving the problems of extracting time series features using the example of EURUSD price dynamics, variational autoencoders have great potential in extracting individual pattern description features.
Author: Dmitriy Gizlyk