Current Solutionís Details View

Online Tutorial - Current Solution's Details

When selected, shows the current solutionís statistics.

Solutionís Statistics

Input Period (days): The number of days or bars the current solution utilizes as a single input.
 
Complexity Level: The complexity level of the current solution. This is an important value because it holds significance in the generalization capability of the solution. The lower the level, the more generalized the solutionís outputs are. The higher the level, the more precise the solutionís outputs are. The difficulty is finding the right balance between these two. Too much precision sacrifices generalization, which tends to make the solutionís predicted outputs unreliable. And to much generalization will make the solutionís predicted outputs inaccurate. The genetic algorithms used to optimize this value, attempts to achieve the perfect balance between these two states.
 
Value Difference: The actual difference in percentage between actual close values and all predicted evaluation values.
 
Correlation %: The correlation percentage ( a statistical measure of similarity ) of all predicted evaluation values vs. actual close values.
 
Survived Generations: When a solution has passed through one generation without being eliminated it has survived a generation. As the system converges, good performing solutions will have survived more generations than poorer performing solutions. This shows the number of survived generations for the given solution. This value will be 0 for solutions that have not been trained yet and are brand new for the population.
 
Comparison to Average: This shows how well this solution compares to the populations average.
 
Error Graph: The lower graph tracks three values:
  • Current Testing Error (Red Line)
     
  • Current Evaluation Error (Green Line)
     
  • Lowest Evaluation Error (Yellow Line)

Professional Only Features

The professional version allows viewing of the internal "Walk Forward" test period by right clicking while on the graph and selecting the "Walk Forward" option. This is a special test that truly shows if the solution has achieved success and is heavily used in Genetic Optimization of the system. It uses the same algorithm in the prediction period where the outputs are wrapped back into the inputs in order to walk-forward (so to speak) into future predicted values. This is how ANNI can produce up to 90 days of future predictions. By using this same algorithm in the evaluation period without the use of actual values, it provides an incredibly rigorous test to determine how well the solution can predict several bars forward by comparing itís results to the actual values. This walk forward period is half of the evaluation period, which can easily be over 30+ bars. For a solution to successfully see the inherent trend over a 30+ bar period without relying on any actual values is astounding. Professional users have the option of viewing this walk-forward period in the output graph for detailed analysis of the best solutions. When active, an additional red line will appear marking the beginning of the internal walk-forward period.

Caution: For optimum training speed, do not keep this option selected. It is a CPU intensive computation and should be used for quick evaluation only.

See Also

What Are Neural Nets

Send feedback about this topic to Oolution Technologies. © 2011 - 2013 Oolution Technologies. All rights reserved.