1. EVA 2.50 Software and Input Values Our paper illustrates the application of a neural network using the Windows version of C#. The training and tests values employed as exemplars within this paper were obtained from the Ninness, Rehfeldt, & Ninness (2019) study. The feedforward backpropagation algorithm that operates within the EVA 2.50 (i.e., EVA 2.5_cn) application was originally developed and described by McCaffrey (2017); also, refer to Haykin (2009) for a discussion. A functional/beta Windows version of EVA 2.50 (with training and test values) is downloadable for academic researchers.

2. The EVA 2.50 application is enclosed within a compressed folder in conjunction with training data and test data files. The current version of EVA 2.50 runs on most Windows operating systems; however, before running the current Windows version of EVA 2.50, some Windows machines may require the installation of a freely available Microsoft program: .
After downloading the compressed folder, it is important to right click and “extract all.” before beginning the installation process. Subsequent to extraction, the user opens the EVA 2.50 folder and clicks the EVA icon. If a Microsoft or other antivirus warning appears (and it will), researchers are advised to read the information below prior to installing and running the program.

3. The current beta version of this application was designed to run on Windows 7 through 10 operating systems (a Mac OS version is in progress). All versions of the EVA 2.50 neural network system are and will remain freely accessible to interested academic users; however, the current beta version of EVA 2.50 is designed for academic research and demonstration purposes only. The authors and journal (The Psychological Record) assume no liability for any damages associated with using, modifying, or distributing this program. The program has not been extensively field tested with input values (datasets) beyond those described in our recent publications; however, researchers are free to employ a variety of training and test values.

4. This program is made available to interested educators and researchers without cost and without any warranties or provision for support. The authors and journal assume no responsibility or liability for the use of the program or provide any certification, license or title under any patent, copyright or government grant. The authors and journal make no representations or assurances with regard to the security, functionality, or other components of the program. There are unidentifiable hazards associated with installing and running any software application, and users/researchers are responsible for determining the extent to which this program is compatible with the computer and other software currently installed on the user’s computer.

5. The EVA 2.50 (EVA 2.5_cn) application is enclosed within a doubly zipped/compressed folder that contains training data and test data files.
The entire zipped folder must be downloaded intact. Subsequent to downloading the zipped folder, the user right clicks the folder and selects “extract all.” Since this is a double folder, there will be a folder inside of the downloaded zipped folder.

6. Upon right clicking and extracting all from the zipped folder, double click the inner folder shown below.

7. Double clicking the EVA2 icon initiates the instillation process.
When the contents of the zipped folder have been opened, the application files will be positioned as shown below.
Double clicking the EVA2 icon initiates the instillation process.
Double click the EVA2 icon

8. An application install Security Warning will appear
An application install Security Warning will appear. In order to install the program, the “install” button must be clicked.

9. At this point, two more Microsoft messages will appear
At this point, two more Microsoft messages will appear. To install the program, the “More info” and “Run anyway” buttons must be clicked one after the other.

10. At the same time, the application is installed on the user’s computer.
Subsequently, the EVA 2.50 artificial neural network application will open.
At the same time, the application is installed on the user’s computer.
See the Appendix with Ninness et al. (2017) for details on conducting a CM Analysis with Training and Test Stimuli. This Appendix can be obtained by clicking the link below:

11. Within the Ninness et al, 2019, Fig. 10. shows the EVA 2
Within the Ninness et al, 2019, Fig shows the EVA 2.50 Windows form permitting interactive training and testing of simulated participants. Training values are shown under the “Input Values” heading. Training Accuracy and Error levels are shown in the center under “Training Outcomes,” and generalization test outcomes are shown under “Test Outcomes.” Momentum is set at 0.38, and the Learning Rate is set to The number of Hidden Layers is 2, and the number of neurons within each of these two hidden layers is 4. See Input Settings in Ninness et al. (2019) for related details.

12. Constructing Models
The remaining slides address saving and reusing weights and bias values that are computed and saved during the training of each simulated participant. It is important to remember that saving and reusing weights in this version of EVA is only possible when the user employs an architecture with one, two, or three layers and the same number of nodes within each of the layers (as described in Ninness et al., 2019).
Five sample CSV weight files (models) are located within the zipped folder; however, these CSV files/models are made available for illustration and discussion purposes only.

13. Constructing Models
During training, the final weights and bias values are automatically saved to the desktop under the heading of “Weights.” These weights must be named after each simulated participant is trained (e.g., Weights_101 for Participant 101). If these files are not individually named after each individual training, the weight and bias values will cumulate within the same CSV file with each newly trained simulated participant.

14. Constructing Models
When a set of weights and bias values have been saved with a specific name, those weights and bias values constitute a “neural network model” based on the training of a particular simulated participant. Again, Weights_101 is a neural network model based on the training of Participant 101.

15. Constructing Models
Individual models may be accessed and used to predict outcomes for entirely new data sets (i.e., test values). When a set of weights and bias values are accessed, they will appear in the middle panel under “Randomized Training.”

16. Constructing Models
Clicking the “Test Weights” button will open a window allowing access to the test values. In this paper we have identified the test values as “Test_22_row_14_cols” since we have 22 rows and 14 cols of test values in this CSV file.

17. Constructing Models
When testing with weights, be sure to use the “Test Weights” button rather than the “Test” button.

18. Constructing Models
After clicking the “Test Weights” button and selecting a training file, the results will appear under “Test Outcomes.”

19. When running EVA 2.50, the user must employ the drop down selection box to obtain a simulation number (seed value). Inserting numbers directly into this location will produce an error message. As shown below, the number 77 was typed into this location and an error value is displayed.

20. Note that our revised version of EVA 2
Note that our revised version of EVA 2.50 is more efficient; however, this version tends to generate slightly higher values for a few of the Related Stimuli (i.e., those values between B1C1 and A2C1) within Experiment For example, we found that our current version of EVA 2.50 produces slightly higher relations at B1B2 than our original outcomes shown below from Experiment 3.
Bear in mind that our revised version of EVA 2.50 is a beta version of the program. As such, it is possible that users may encounter occasional malfunctions and discrepancies within their findings. Thus far, we have found discrepancies to be minimal. As indicated above, the program has not been extensively field tested with input values (datasets) beyond those described in our recent publications.

21. Slide References
Haykin, S. O. (2009). Neural networks and learning machines (3rd ed.). Upper Saddle River, NJ: Pearson Education.
McCaffrey, J. (2015). Coding neural network back-propagation using C#. Retrieved from
McCaffrey, J. (2017). Test run: Deep neural network training. Retrieved
from
Ninness, C., Rehfeldt, R. A. & Ninness, S. (2019) Identifying accurate and inaccurate stimulus relations: human and computer learning. The Psychological Record. see
Ninness, C., Ninness, S., Rumph, M. & Lawson, D.  (2018). The emergence of stimulus relations: human and computer learning. Perspectives on Behavioral Science, 41, 121–154. Initially published online November 2017 in The Behavior Analyst. see