compat.PyTorch2RiverClassifier problem

[ghost]

Hi!

I am struggling with compat.PyTorch2RiverClassifier, (A river classifier that integrates neural Networks from PyTorch) could anyone help me please?
I have a splipt with my resnet18 model and its pre-trained weights in pytorch and i would like to use river, but it seems like is impossible to make an inference! I define the class model as always in pytorch, loaded weights but i dont know how to use:

class definition
class Resnet18_Model(torch.nn.Module):
def init(self):
super().init()
self.model=models.resnet18(pretrained=True)
for param in self.model.parameters():
param.requires_grad = False
num_ftrs = self.model.fc.in_features
self.model.fc=nn.Sequential
nn.Linear(num_ftrs, 2), #2 possible output
nn.Softmax(dim=1)
)
def forward(self, xb):
return self.model(xb)

def build_classifier(n_features):
return Resnet18_Model()

pytorch to river conversion
rivermodel = river.compat.PyTorch2RiverClassifier(build_fn = build_classifier,
loss_fn = nn.CrossEntropyLoss)

image inference
rivermodel.predict_proba_one(image)

Image is my tensor (3d image) but i dont know how to give it to the predict_proba_one function. I can read in docs that it must be a dict() but how? Should i flatten the 3d tensor?

Where shoud i instance the model? (model=Resnet18_model())?

Any help would be appreciated.

[MaxHalford]

Could you check out river-torch? We're going to deprecate the PyTorch stuff in the compat module.

[ghost]

Hi @MaxHalford, thank you for your help, but i have the same problem here,....how to send my 3d tensor image to the predict_one(x) function? x must be a dict, right?

Thank you

[MaxHalford]

Yes we only work with dictionaries, not tensors.

[ghost]

OK, then I assume that this library is not designed to work with images (tensors), as we can do in pytorch. As I see this is more oriented to ML aplications instead of DL aplications and thats why your example datasets are in .csv format. right? If don't, I would appreciate if you can share an example in which I could see how to make inference with a tensor.

Thanks for your support.

[MaxHalford]

Indeed we cater to classic ML problems :)