@formula for Lathe.preprocess UndefVarError

Specific Domains Statistics
1

Hello Everyone:

I am attempting to perform a logistic regression and
for some reason the @formula macro, is not reading.


using GLM
using Lathe: preprocess


fm = @formula(Outcome ~  Sesh1 + Shes2 + Sesh3)
logit = glm(fm, train, Binomial(), ProbitLink())

I attempted to prefix @formula and there are
no package conflict warnings. One thing, I have
been getting some deprecation warnings during
my Lathe install. But for other sessions I have
been able to use Lathe package and modules.

Any suggestions?

2

Could you turn this into an MWE? Or at a minimum say what you mean by “the formula macro is not reading”?

3

Sure

I am working with the following variables:

ST = ["Sesh1","Sesh2","Sesh3"]
RT = [1,2,3]
Avail = ["Y", "N"]

The data frame I am working with is is structure similar to:

DBH = DataFrame(Id = 1:400, 
		    SessionTypes = rand(ST, 400),
		     RequestTypes = rand(RT, 400), 
		     Available = rand(Avail, 400))

Then am encountering the error when:

fm = @formula(Available ~  SessionTypes + RequestTypes)
logit = glm(fm, train, Binomial(), ProbitLink())
4

Okay first of all it’d be good if you had a look at Please read: make it easier to help you to get some tips on how to most effectively ask for help on here - your example isn’t exactly minimal I suppose if your error is related to @formula. Also, what is the actual error you’re seeing?

5

Your problem is that DBH.Available is a vector of Strings, not booleans. You can do

DBH.Available = DBH.Available .== "Y"

and your code will work.

GLM should definitely throw a better error in this scenario, though.

6

Thank you @pdeffebach

Yes – I pre-fixed @ formula before your
suggestion and received a similar error
message. Am waiting for my REPL to
load up in my browser.

Would you apply these instructions in the
first block as:

Avail.== "Y"

OR

Avail .== ["Y","N"]

If so, the output will vector will auto-assign
“Y” as 1 or else 0?

Or block two as:

DBH = DataFrame(Id = 1:400, 
		    SessionTypes = rand(ST, 400),
		     RequestTypes = rand(RT, 400), 
		    Available = rand(Available .== "Y")

I want to make sure the DBH.Available vectors are random.

Thank you,

7

Definitely this one. The 2nd one will throw an error (which you will soon realize as you try running the command…)

You can also do it on the DataFrame construction.

8

@pdeffebach Appreciated.

What is the logic behind this, how
does the system know to treat “Y”
as a bool? Or some other wizardry
at play?

9

It sounds like you could use a tutorial. or read through the Julia docs. It’s just checking for equality with the string "Y", which returns a Bool, as it does in all other languages.

Please read the documentation so you have a better idea what’s going on!

1 Like
10

Okay.

From your review, which you may
have gotten from HERE the instantiation
and data frame construction should
involve these steps:

ST = ["Sesh1","Sesh2","Sesh3"]
RT = [1,2,3]
Avail .== "Y"

Then,

DBH = DataFrame(Id = 1:400, 
		    SessionTypes = rand(ST, 400),
		     RequestTypes = rand(RT, 400), 
		     Available = rand(Avail, 400))

The output should replace the “Y” with 1 and
any other value with 0.

I am still wondering, after reviewing the docs, at
least on Strings… how the 0 is assigned randomly.
We did not in this example instantiate “N”.

11 
Avail .== "Y"

Again you need to try your code out. This will result in an error without Avail being previously defined.

12

@pdeffebach Understood.

The way you worded what you said previously,
made me think in the FIRST block you were
instantiating Avail as:

Avail .=="Y"

Instead of:

Avail = ["Y","N"]

Then I realized you meant to apply
Avail .==“Y” to the SECOND block as:

DBH = DataFrame(Id = 1:400, 
		    SessionTypes = rand(ST, 400),
		     RequestTypes = rand(RT, 400), 
		    Available = rand(Available .== "Y", 400)

Now I understand why the equality works and
before thought it was some obscure instructions.

Thanks.

13

The error read:

no method matching fit(::Type{GLM.GeneralizedLinearModel}, ::Matrix{Float64}, ::Matrix{Float64}, ::Distributions.Binomial{Float64}, ::GLM.ProbitLink)