I am not able to understand the part at 1 hour 22 minutes of lecture 3. As Jeremy said that we don’t need a third dummy variable for 3 categorical variables as it can be done without that. But my question is how??
I mean as I have read till now for 3 categorical variables we do create 3 dummy variables as we do in encoding.
Can someone explain this, please?
Hello,
Pclass can take on three values, namely, 1, 2, or 3. Pclass_1 is 1 if Pclass is 1 and 0 otherwise. Similarly, Pclass_2 is 1 if Pclass is 2 and 0 otherwise. Therefore, by definition, when Pclass_1 and Pclass_2 are both zero, it immediately follows that Pclass is neither 1 nor 2 and is thus 3. In other words, a hypothetical Pclass_3 would be superfluous since it is 1 if and only if Pclass_1 and Pclass_2 are 0 and 0 otherwise.
Below is a table of the possible combinations of Pclass, Pclass_1, and Pclass_2. Note that Pclass_1 and Pclass_2 cannot both be 1 because that would imply Pclass is 1 and 2 simultaneously.
--------------------------------
| Pclass | Pclass_1 | Pclass_2 |
--------------------------------
| 1 | 1 | 0 |
| 2 | 0 | 1 |
| 3 | 0 | 0 |
--------------------------------
Does that clarify the matter?
Hello
This may help or totally confuse. We normally count in 10 so 57 is really 510+7 but we could count in 2 so 57 = 32 + 16 + 8 + 1 or 22222 + 2222 + 222 + 1. We could write this as 1 1 1 0 0 1. This is called binary. Now consider the number 3 as in 3 classes. We could write class 1 as 20+1, class 2 as 12+0 and class 3 as 1*2 +1. Hence C1 = 01, C2 = 10 and C3 = 11. Whether C3 = 11 or 00 is just a personal choice a more pragmatic approach would be to use class -1 to be 00 01 10 respectively. This would be consistent if you had a larger number of classes.
Regards Conwyn