[Vorbb-tech-devs] Matrix Movie Results

[Brian C. B. <bri...@gm...>]

Hey Jason/all,

Some interesting things. I think a trend is beginning to emerge if you go
with movies instead of customers. Playing around with those error files from
your method I noticed some cool stuff:

- If you could eliminate error on the 10 worst movies (total error, not
normalized), you could improve your RMSE by 0.05
- Worst movies seem to pretty recent (Napolean Dynamite, Spotless Mind,
etc), probably because they have many more ratings
- Worst movies actually have a VERY normal distribution, only flattened
(i.e. uniform with a bulge in the middle)
- Best movies have a high skewed to the right distribution, almost looks
like an exponential curve. 

Anyhow, wanted to let you know. We can talk about what it means later. 

Sincerely,
Brian C. Becker @ www.BrianCBecker.com