iobt-vistec / min2net Goto Github PK

I would like to use openBCI where MI-EEG can get 16 channels at 125hz. In this case, how should I change the input and model of MIN2Net?

Since binary f1-score consider precision and recall, it does not consider TN.
Let me consider right hand class as the positive and left hand class as the negative. For this case, binary f1-score ignores left hand class.
Conversely, when the positive and negative labels are changed, the f1-score is different.
Because the binary MI classification is not a positive-negative problem, we should treat both class equally.
Therefore I suggest use macro f1-score for the binary MI classification.

Does MIN2Net do the classification like encoder->decoder->softmax?
My understanding of MIN2Net without decoder is that the structure is encoder->softmax.

Due to TensorFlow's compilation issues, the second loop attempts to use the first initialize object of the optimizer in the __init__ function as the loss return NaN value (Even if the new model object is declared within the loop, the first initialize optimizer is attempted)

To resolve this issue, do not utilize the optimizer's initial value as provided by the __init__ function; instead, the optimizer must be assigned in each and every fold as in the example below:

for fold in range(1, 6):
    # the optimizer must be assigned in each and every fold
    optimizer = Adam(beta_1=0.9, beta_2=0.999, epsilon=1e-08)
    model = MIN2Net(optimizer =optimizer)

However, this issue does not affect the experimental results of our paper since we declared a new optimizer in every loop https://github.com/IoBT-VISTEC/MIN2Net/blob/main/experiments/train_MIN2Net_k-fold-CV.py#L41

Can you share the code of t-SNE?

I was able to download the BCIC2a raw data with min2net.utils.load_raw, but how do I download the time_domain?

Only the MI-EEG classification part was expressed in pytorch. Does it match?

class Conv2D_Norm_Constrained(nn.Conv2d):
    def __init__(self, max_norm_val, norm_dim, **kwargs):
        super().__init__(**kwargs)
        self.max_norm_val = max_norm_val
        self.norm_dim = norm_dim

    def get_constrained_weights(self, epsilon=1e-8):
        norm = self.weight.norm(2, dim=self.norm_dim, keepdim=True)
        return self.weight * (torch.clamp(norm, 0, self.max_norm_val) / (norm + epsilon))

    def forward(self, input):
        return F.conv2d(input, self.get_constrained_weights(), self.bias, self.stride, self.padding, self.dilation, self.groups)

class ConstrainedLinear(nn.Linear):
    def forward(self, input):
        return F.linear(input, self.weight.clamp(min=-1.0, max=0.5), self.bias)
class MinNet(nn.Module): # input = (1,16,125)
  def __init__(self, input_shape=(1,400,20)):
    super().__init__()
    self.D, self.T, self.C = input_shape
    self.subsampling_size = 100
    self.pool_size_1 = (1,self.T//self.subsampling_size)
    self.en_conv = nn.Sequential(
                    Conv2D_Norm_Constrained(in_channels=1, out_channels=16, kernel_size=(1, 64), padding="same", max_norm_val=2.0, norm_dim=(0, 1, 2)),
                    nn.ELU(),
                    nn.BatchNorm2d(16,eps=1e-05, momentum=0.1),
                    nn.AvgPool2d((1,self.pool_size_1)),
                    nn.Flatten(),
                    ConstrainedLinear(32000,64),
                    nn.ELU(),
                    ConstrainedLinear(64,3)
                )
  def forward(self,x):
      x = self.en_conv(x)
      return x

What is self.subsampling_size?
The self.subsampling_size in the MIN2Net model is 100, what is the value?

Google colab occurs error.
Collecting min2net
Using cached min2net-1.0.1-py3-none-any.whl (58 kB)
Using cached min2net-1.0.0-py3-none-any.whl (57 kB)
ERROR: Cannot install min2net==1.0.0 and min2net==1.0.1 because these package versions have conflicting dependencies.

The conflict is caused by:
min2net 1.0.1 depends on tensorflow-addons==0.9.1
min2net 1.0.0 depends on tensorflow-addons==0.9.1

To fix this you could try to:

1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict
   But google colab only use tensorflow-addons >= 0.10.0.
   How I do solve this?