Cross-validation Train MSE

[hlmtran]

Even with train_track_loss flag set to false, it seems to report mse for runs that reach max iteration. This looks to be separate from HECA and MOCA decompositions having 0 MSE for train set loss during CV.

library(RcppML)
data(aml)

testCV = RcppML::nmf(aml,k=c(2:10),test_fraction = 0.05,track_train_loss=F,seed=12,cv_seed=123)

testCV2 = RcppML::nmf(aml,k=c(2:10),test_fraction = 0.05,track_train_loss=F,cv_seed=125)

ggplot(testCV, aes(x = k, y = train_mse)) +
  geom_line(color = "grey60", linewidth = 0.6) +
  geom_point(aes(color = total_iter), size = 1.5) +
  scale_color_viridis_c(
    option = "B",
    limits = c(0, 100)
  ) +
  labs(
    title = "seed 123, track_train_loss=F",
    x = "Rank (k)",
    y = "MSE",
    color = "Iteration"
  ) +
  theme_minimal()

ggplot(testCV2, aes(x = k, y = train_mse)) +
  geom_line(color = "grey60", linewidth = 0.6) +
  geom_point(aes(color = total_iter), size = 1.5) +
  scale_color_viridis_c(
    option = "B",
    limits = c(0, 100)
  ) +
  labs(
    title = "seed 125, track_train_loss=F",
    x = "Rank (k)",
    y = "MSE",
    color = "Iteration"
  ) +
  theme_minimal()