fangwei123456

> Is it same as training pre-trained models in pytorch? Yes, you build a custom dataset, and then you can train your SNN on this dataset. It is same with...

> Is it possible to use spiking LSTM from spiking jelly to classify iris dataset?. Yes, iris is a simple dataset, and I think spiking LSTM can handle it.

> Do I need to replace IF with LIF Yes, you only need to replace it: https://github.com/fangwei123456/spikingjelly/blob/1f01c9c500f73e0814cdc0dca4fb2f9ac81ad57a/spikingjelly/clock_driven/examples/cifar10_r11_enabling_spikebased_backpropagation.py#L124 But notice that this network is special because its backward calculation is different...

`save_max` is not used.

If you want to save the model with maximum acc, you can do like this: ```python if test_acc > max_test_acc: max_test_acc = test_acc PATH = "/mnt/beegfs/home/sgilani2020/saved models/state_dict_resnet18_pre_trained1.pt" print(max_test_acc) torch.save(net.state_dict(), PATH)...

IRIS dataset only has three classes. Does the output of SNN you used have shape [batch_size, 3]?

The default classes_num is 1000: https://github.com/fangwei123456/spikingjelly/blob/72132ff2d147ceaefe1f68aa74d6286086c62750/spikingjelly/clock_driven/model/spiking_resnet.py#L184

My email is [email protected]. But I will not reply quickly because I am busy in doing other things.

A larger/deep network does not guarantee a better accuracy. You can check the training accuracy of two networks. If the training accuracy of spiking resnet-18 is lower than spiking VGG-6,...

In this tutorial, the input has `shape = [N, 1, H, W]`, and the first conv layer is `c_in=1, c_out=128`. We will get the output spikes with `shape = [N,...