1. Optimizer

1.1. Introduction

The optimizer is a plug-in that uses artificial intelligence algorithms to optimize rules, ranking systems, trading systems and prediction items. 

1.2. Optimizer algorithms

The plug-in provides two algorithms:

Example:

Take a list of rules with 1000 rules.
For these rules, we have more than billions of combination which will take years to back-test.
Optimization algorithms are used to find good solutions among all the combinations in a short period.

1.2.1. Genetic algorithm

These are the genetic algorithm parameters:

• Number of generation
• Population size
• Selection method: Elite, Rank or Roulette
• Crossover rate
• Mutation rate
• Random selection portion

http://en.wikipedia.org/wiki/Genetic_algorithm

1.2.2. PBIL algorithm

These are the PBIL algorithm parameters:

• Number of generation
• Population size
• Learning rate
• Number of best solutions to use in learning

http://en.wikipedia.org/wiki/Population-based_incremental_learning

1.3. Optimize items

1.3.1. Optimize list of rules

Search for the best combination of rules among a list of rules. 

• In the 'Optimizer' form, click on 'Create'
• Select the optimization algorithm and 'List of Rules' in the 'What do you want to optimize?" list box
• Click on 'Next'
• Update the algorithm settings
• Click on 'Next'
• Select the number of rules
• Select a list of rules, by clicking on 'Select rules'.
• For list of rules, select whether to always include the current rule or not by checking or un-checking 'Always visible' checkbox
• Select the output
• Define a fitness formula
• Click on 'Next'
• Select the number of items to keep in the report
• Select the symbols the optimizer will use
• Click on 'Next'
• The optimizer item will now appears in the optimize manager.

Example:
List of rules contains 100 rules, from rule1 to rule100.

The optimizer will try to find good combination of rules.
The result will look like this:

Fitness could be the return of the symbol or something else; it depends on the 'Fitness' formula.

1.3.2. Optimize ranking system

Try to find which combination of nodes and nodes' weight gives the best fitness score for the ranking system.

• In the 'Optimizer' form, click on 'Create'
• Select the optimization algorithm and 'Ranking System' in the 'What do you want to optimize?" list box
• Click on 'Next'
• Update the algorithm settings
• Click on 'Next'
• Select a ranking system by clicking on 'Select a ranking system'
• Select the number of buckets and the rebalance period
• Choose whether to optimize nodes' weight by checking or un-checking 'Optimize node weights' checkbox
• Type a fixed rule if you want to use a fixed rule in the ranking system evaluation
• Select the output
• Define a fitness formula
• Click on 'Next'
• Select the number of items to keep in the report
• Select the symbols the optimizer will use
• Click on 'Next'
• The optimizer item will now appears in the optimize manager.

1.3.3. Optimize trading system
 

Try to find which trading system works best given a buy, sell, short and cover list of rules, money management variables, and others settings. 

• In the 'Optimizer' form, click on 'Create'
• Select the optimization algorithm and 'Trading System' in the 'What do you want to optimize?" list box
• Click on 'Next'
• Update the algorithm settings
• Click on 'Next'
• Select the number of buy and sell rules
• For each buy and sell rule, select a list of rules by clicking on 'Select rules'
• For each buy and sell rule, select whether to always include this rule or not by checking or un-checking 'Always visible' checkbox
• Define a fitness formula
• On the 'Settings' tab, type a fixed buy rule and a fixed sell rule if you want to include fixed rules in your trading system
• Select 'Optimization', to include capital and stops optimizable variables
• Click on 'Next'
• Select the number of items to keep in the report
• Select the symbols the optimizer will use
• Click on 'Next'
• The optimizer item will now appears in the optimize manager.

1.3.4. Optimize prediction item
 

Try to find which prediction item leads to a better prediction accuracy given a list of inputs.

• In the 'Optimizer' form, click on 'Create'
• Select the optimization algorithm and 'Prediction Model' in the 'What do you want to optimize?" list box
• Click on 'Next'
• Update the algorithm settings
• Click on 'Next'
• Select a prediction item then click on 'Load Inputs'
• Update the prediction item if necessary by clicking on 'Update Item'
• Select the input that will always be used in iterations by checking them (Fixed inputs)
• Define a fitness formula
• Click on 'Next'
• Select the number of items to keep in the report
• Select the symbols the optimizer will use
• Click on 'Next'
• The optimizer item will now appears in the optimize manager.

1.4. Optimize manager

The 'Optimize manager' list all optimize items (AI -> Optimizer)

To create an optimize item, click on 'Create'.
To remove an optimize item, click on 'Delete'.
To update an optimize item, click on 'Update'.

To load the optimizer report, click on 'Report'.
To start an optimize item, click either on 'Run' or on 'Run...'

Running an optimize item indefinitely, using 'Run...', means that when the optimizer reach the maximum number of generation or the fitness value converge, the software re-run the optimize item and increase the value of 'Turn' by one.
 

1.5. Turn

If you run an optimize item multiple times, you may end up each time with different results.
This is because the optimizer algorithms do not look for all combinations but instead try to find good solutions in a short period.
This is why we introduced the concept of 'Turn'.
The optimizer report includes the best items for all the turns (not only the last turn)..