AIsolver

AI SOLVER STUDIO WALK-THROUGH

Prerequisites:

As AI Solver Studio attempts to solve classification problems, you need to have a specific problem you would like to solve and data for that problem. Alternatively you can use problems from the SAMPLE directory.
To use the program you will need to have a ready data file. AI Solver accepts delimeted text files (such as CSV). To learn more about the structure of AI Solver data files please study the files in the SAMPLES directory.

At a minimum a single data file as needed (we will call this the training data set). This is the data AI Solver will use for learning. It is highly recommended to use a secondary data file (testing data set) to validate the learning performed. As a rule of thumb, the testing data set should be about 1/4 – 1/3 of the training data set, although this recommendation is not enforced.

This walk-through also assumes you have already installed and started the program.

STEP #0: Settings

To make sure the settings of AI Solver fit your configuration, click Options->Performance.
Select the number of cores your CPU has and the priority given to worker threads.
If you are not sure how many cores your CPU has, leave the setting at single core. Normal priority is recommended for most users, while lowest is recommended if your computer is running other critical tasks at the same time as AI Solver. Highest priority is recommended if you ar e running AI Solver on a dedicated computer doing little or nothing else.

STEP #1: Create a project

Now, let’s create your first project.
Click File->New Project.

STEP #1a: Select project type

Select what kind of project you would like to create. The options are:

• Neural network project
• Genetic programming project
• Combo project

This selection determines the constructs used for learning. Neural networks and genetic programming have a large body of available literature on the web and are not covered in this document. If you are unsure or have no preferences regarding this, it is recommended to start with a neural network project.

STEP #1b: Import training data

Type in the path or browse for your training data file. If you are not using semicolon as your delimiter symbol, change this as well. If the import of your data fails for some reason, you will be notified immediately.

STEP #1c: Error severity handling

In this step you choose what kind of error severity handling you would like for your project. If you are dealing with a problem where all errors (misclassifications) are equally serious, select the upper option. If however you would like different levels of severity for different kinds of errors, select the lower one.
If the latter is selected, you will next be shown a window in which you can specify the error severity for different kinds of errors. There are four levels of error severity with self-explanatory naming. These settings will influence the learning process which prioritize itself in order to avoid the most serious types of error.

STEP #1d: Hindsight

Hindsight is a special feature in AI Solver intended to be used with time series data where needed. As time series data can in general by viewed as a list of events, hindsight represents the ability to look beyond the current event to previous events.

If you enable hindsight, you need to select how many rows (events) from the past will be seen. A value of one indicates that the current event and the one before it will be used as inputs to the solution.
Please note that hindsight and increasing hindsight values increase the complexity of your solution and thus the needed training time. It is recommended to only use this feature if you need it.

STEP #1e: Solution complexity

This step allows you to select the desired complexity for your solution. This selection affects a number of internal parameters of the system. It is recommended to start with a complexity setting of Simple and increase complexity if acceptable performance is not attained in the learning process.

STEP #1f: Overfitting prevention

Overfitting prevention is a feature in AI Solver that is useful for preventing over-generalization from patterns occurring in the training data that do not have general use. If this option is selected, you need to select how much of the training data you will allocate to this feature. Keep in mind that this effectively reduces the training data set. As with Hindsight, it is recommended to use this feature only if you encounter overfitting problems in your solutions.

This concludes the project creation.

STEP #2: Load testing data

To load testing data for your project, click Data->Load testing data.
Type in the path or browse for your testing data file.

STEP #3: Start training

Now that your project has been created and necessary data imported you are ready to start the training process. Simply click the Start training button.

Throughout the training process, the main window will continously show various information such as training progress and performance statistics. If your problem is fully solved the training will come to an end automatically with a message to indicate this. More commonly the user will need to evaluate the performance and manually stop the training process once acceptable results are achieved.

To stop the training, click the Stop training button.

STEP #4: Processing unclassified data (optional)

AI Solver provides the ability to process unclassified data, i.e. new data not seen in the training or testing data sets.

To accomplish this, click Data->Process unclassified data and type in the path or browse for your data file.
Note that an unclassified data file has one fewer column than training or testing data files as the result column is not present.

Popularity: 12% [?]