When sample size converge into reads

When we look at the stats V$IP and PFR we can usually get a read on a villain after 25 to 50 hands. Of course that read becomes more reliable with more hands tracked. I used to label players after 50 hands or so based on these 2 stats but lately I am starting to think that is not a good idea. Instead of trying to categorize players like I used to, I am now expanding by decisions by including the following stats.

3bet PF
Fold to 3bet
Cold call
Fold To Cbet

The problem is everytime I sit down for a session I have no hands on about half the table. And even if I do have their Fold To CBet stat its usually along the lines of 2/5 or 3/4. So for the stat heads out there, what are some reasonable sample sizes for these stats? Hopefully we can come up with a nice list here so if anyone wants a stat to be included, just reply and I will post it up here. (BTW I did a search with no results if there is a thread like this already I apologize)

Taking a wild stab I will venture to say the following number of hands are required to base any type of decision on.

3Bet PF - 50 hands or so. I would stay away from a 2/67 and get it in vs 8/45
Fold to 3Bet - About 10 hands.
Cold call - Not gonna even attempt this one
Fold to Cbet - 25 hands

can you explain what does mean Cold call???

Calling (but not raising) when faced with more than one bet in a single action.
If one opponent bets, and another raises, and you call, you have cold called.