beta0 should be computed using m, not n

surajiyer / Transfer-learning-with-TrAdaBoost

Scikit-learn style implementation of TrAdaBoost algorithm

36 stars 11 forks source link

beta0 should be computed using m, not n #1

Open KKIverson opened 3 years ago

KKIverson commented 3 years ago

TrAdaboost-alg according to the paper, beta0 should be computed using m (the number of different samples）

DavideNardone commented 3 years ago

For the pseudo-code provided above, I see that the equation accounting for beta0 is computed using n. Why are you stating differently?

KKIverson commented 3 years ago

For the pseudo-code provided above, I see that the equation accounting for beta0 is computed using n. Why are you stating differently?

Because in the pseudo-code n is the number of samples in target domain and m is the number of samples in source domain, beta0 is computed using n. But things are different in the source code, n and m represent the number of samples in the source and target domains, respectively. Beta0 should be always computed using the number of samples in target domain.

sucre111 commented 3 years ago

For the pseudo-code provided above, I see that the equation accounting for beta0 is computed using n. Why are you stating differently?

Because in the pseudo-code n is the number of samples in target domain and m is the number of samples in source domain, beta0 is computed using n. But things are different in the source code, n and m represent the number of samples in the source and target domains, respectively. Beta0 should be always computed using the number of samples in target domain.