dragen1860
Testing-Strategy
Hello, Thanks for the code. I have a question regarding testing strategy. I trained model and run the with --test parameter.
During test time I can see from the output that something like that: [support_t0, query_t0 - K] mean: [ 0.20175 0.37785035 0.41440004 0.43647221 0.44063893 0.44136062 0.44179988 0.44246089 0.44228852 0.44238317 0.44273853] .....
Here querry results are trained model accuracies. We are using query_y to refine our prediction and computing the accuracy for 10 iteration. This approach doesn't make sense to me in a few shot learning setting. During test time, I was expecting that we will train model on K samples from classes [C1,C2, ..,Cn] and test on the other samples from these classes [C1,C2, .., Cn]. Here we are training model still in all test samples. update
