Low accuracy about bindsnet

[Hananel-Hazan]

Oh ok, please use eth_mnist.py. The script you use batch_eth_mnist.py is meant to show how to use the batch method and this script is not optimized for the best performance.

[Hananel-Hazan]

Try to use the default parameters.

[bsar321]

Hi,
With default parameters the accuracy obtained is 80% which is very low compared with the accuracy of paper.
Can you help me ?

[Hananel-Hazan]

Which paper are you referring to?

[bsar321]

I reffering to "BindsNET: A Machine Learning-Oriented Spiking Neural Networks Library in Python" , in section 4 (examples) more specific in subsection 4.1 is mentioned an accuracy of 95% but i can't get these accuracy.

[bsar321]

Thank you, but before I used eth_mnist.py and the precision was 80% with the default parameters, what do I need to change to get the 95% ? thanks

[Hananel-Hazan]

The script used to reach 95% is SOM_LM-SNNs.py which is mentioned here Unsupervised learning with self-organizing spiking neural networks

[bsar321]

Ok, i will try, do you recommend using the default parameter or others to obtain 95%? Thanks again

[Hananel-Hazan]

I think the default parameters should get you higher accuracy then 80%. In the paper there is a table of the number of neurons and the their accuracy. I think the script should give you close enough to the paper. There are many changes in BindsNET since then, so your mileage may vary :-)