I get the error: heatmap=True
requires a model like cnn_learner
produces.'
When I try to call
interp.plot_top_losses(9, figsize = (8,6), heatmap = True, largest=top).
I understand this is because I initialized my own custom Learner instead of cnn_learner(I used my own model)
learn = Learner(data, get_cifar_model(), loss_func=nn.CrossEntropyLoss(), metrics=[accuracy])
But when I initialize it with cnn_learner I get the exact error described here:
“Exception: No weight layer”
Is there anyway to get heatmaps from plot_top_losses with a custom model?
Thank you in advance