But, an accurate forward map would require a very large size of training data as well as network architecture.
A reduction in the size of training sets also allows the Levenberg Marquardt algorithm to be implemented.
We then extend our approach to determine the size of training set given a tolerance of approximation error.
We found that the level of accuracy with the size of training set varies for different algorithms.
We used a known learning-curve and introduce a measure named "neuron-curve" to find the optimal layer size & minimum size of training set.
A training data generation method is introduced, which reduces the size of training data by generating the states near the hyperplane.
In these general estimators/regression function/classifiers, parameters to be tuned are the number of layers, neurons, type of transfer function, minimum size of training set, etc.
However, it is subjected to the model selection and learning complexity O K * N3), especially for a massive data set (N is the size of training dataset, and K is the number of search).
In addition, these comprehensive experiments cover most of the current trading system topics, including the use of a sliding window method (with or without validation dataset), the number of trading rules, and the size of training period.
This analysis shows that, as long as few data are available to train the model, splines approaches lead to best results, while when the size of training dataset increases, SVM and RF provide faster and more accurate solutions.
The final size of training data is 8 GB.
