Skip to main content

probability or statistics - Multinomial logistic regression


Has anyone done multinomial logistic regression in Mathematica?


The binomial case is essentially done on the LogitModelFit documentation page and works fine.


I am following this formulation to obtain the outcome probabilities for k classes with k-1 calls to LogitModelFit.


I am getting normalized probabilities out, but they depend on how the classes are encoded for each independent regression. I can post what I have on request but it's a lot of code for something that isn't working.


Either multiple-binary or true multinomial would be fine for my purposes, but bonus for both implementations.



Answer



Multiple-Binary Approach


Lets bring out the famous iris data set to set this up and split it into 80% training and 20% testing data to be used in fitting the model.


iris = ExampleData[{"Statistics", "FisherIris"}];

n = Length[iris]

rs = RandomSample[Range[n]];
cut = Ceiling[.8 n];
train = iris[[rs[[1 ;; cut]]]];
test = iris[[rs[[cut + 1 ;;]]]];

speciesList = Union[iris[[All,-1]]];

First we need to run a separate regression for each species of iris. I'm creating an encoder for the response for this purpose which gives 1 if the response matches a particular species and 0 otherwise.



encodeResponse[data_, species_] := Block[{resp},
  resp = data[[All, -1]];
  Join[data[[All, 1 ;; -2]]\[Transpose], {Boole[# === species] & /@ 
      resp}]\[Transpose]
  ]

Now we fit a model for each of the three species using this encoded training data.


logmods = 
  Table[LogitModelFit[encodeResponse[train, i], Array[x, 4], 
    Array[x, 4]], {i, speciesList}];


Each model can be used to estimate a probability that a particular set of features yields its class. For classification, we simply let them all "vote" and pick the category with highest probability.


mlogitClassify[mods_, speciesList_][x_] :=
 Block[{p},
  p = #[Sequence @@ x] & /@ mods;
  speciesList[[Ordering[p, -1][[1]]]]
  ]

So how well did this perform? In this case we get 14 out of 15 correct on the testing data (pretty good!).


class = mlogitClassify[logmods, speciesList][#] & /@ test[[All, 1 ;; -2]];


Mean[Boole[Equal @@@ Thread[{class, test[[All, -1]]}]]]

(* 14/15 *)

It is interesting to see what this classifier actually does visually. To do so, I'll reduce the number of variables to 2.


logmods2 = 
  Table[LogitModelFit[encodeResponse[train[[All, {3, 4, 5}]], i], 
    Array[x, 2], Array[x, 2]], {i, speciesList}];


Show[Table[
  RegionPlot[
   mlogitClassify[logmods2, speciesList][{x1, x2}] === i, {x1, .5, 
    7}, {x2, 0, 2.5}, 
   PlotStyle -> 
    Directive[Opacity[.25], 
     Switch[i, "setosa", Red, "virginica", Green, "versicolor", 
      Blue]]], {i, {"setosa", "virginica", "versicolor"}}], 
 ListPlot[Table[
   Tooltip[Pick[iris[[All, {3, 4}]], iris[[All, -1]], i], 

    i], {i, {"setosa", "virginica", "versicolor"}}], 
  PlotStyle -> {Red, Darker[Green], Blue}]]

enter image description here


The decision boundaries are more interesting if we allow more flexibility in the basis functions. Here I allow all of the possible quadratic terms.


logmods2 = 
  Table[LogitModelFit[
    encodeResponse[train[[All, {3, 4, 5}]], i], {1, x[1], x[2], 
     x[1]^2, x[2]^2, x[1] x[2]}, Array[x, 2]], {i, speciesList}];


Show[Table[
  RegionPlot[
   mlogitClassify[logmods2, speciesList][{x1, x2}] === i, {x1, .5, 
    7}, {x2, 0, 2.5}, 
   PlotStyle -> 
    Directive[Opacity[.25], 
     Switch[i, "setosa", Red, "virginica", Green, "versicolor", 
      Blue]], PlotPoints -> 100], {i, {"setosa", "virginica", 
    "versicolor"}}], 
 ListPlot[Table[

   Tooltip[Pick[iris[[All, {3, 4}]], iris[[All, -1]], i], 
    i], {i, {"setosa", "virginica", "versicolor"}}], 
  PlotStyle -> {Red, Darker[Green], Blue}]]

enter image description here


Comments

Post a Comment

Popular posts from this blog

functions - Get leading series expansion term?

Given a function f[x] , I would like to have a function leadingSeries that returns just the leading term in the series around x=0 . For example: leadingSeries[(1/x + 2)/(4 + 1/x^2 + x)] x and leadingSeries[(1/x + 2 + (1 - 1/x^3)/4)/(4 + x)] -(1/(16 x^3)) Is there such a function in Mathematica? Or maybe one can implement it efficiently? EDIT I finally went with the following implementation, based on Carl Woll 's answer: lds[ex_,x_]:=( (ex/.x->(x+O[x]^2))/.SeriesData[U_,Z_,L_List,Mi_,Ma_,De_]:>SeriesData[U,Z,{L[[1]]},Mi,Mi+1,De]//Quiet//Normal) The advantage is, that this one also properly works with functions whose leading term is a constant: lds[Exp[x],x] 1 Answer Update 1 Updated to eliminate SeriesData and to not return additional terms Perhaps you could use: leadingSeries[expr_, x_] := Normal[expr /. x->(x+O[x]^2) /. a_List :> Take[a, 1]] Then for your examples: leadingSeries[(1/x + 2)/(4 + 1/x^2 + x), x] leadingSeries[Exp[x], x] leadingSeries[(1/x + 2 + (1 - 1/x...
Post a Comment
Read more

mathematical optimization - Minimizing using indices, error: Part::pkspec1: The expression cannot be used as a part specification

I want to use Minimize where the variables to minimize are indices pointing into an array. Here a MWE that hopefully shows what my problem is. vars = u@# & /@ Range[3]; cons = Flatten@ { Table[(u[j] != #) & /@ vars[[j + 1 ;; -1]], {j, 1, 3 - 1}], 1 vec1 = {1, 2, 3}; vec2 = {1, 2, 3}; Minimize[{Total@((vec1[[#]] - vec2[[u[#]]])^2 & /@ Range[1, 3]), cons}, vars, Integers] The error I get: Part::pkspec1: The expression u[1] cannot be used as a part specification. >> Answer Ok, it seems that one can get around Mathematica trying to evaluate vec2[[u[1]]] too early by using the function Indexed[vec2,u[1]] . The working MWE would then look like the following: vars = u@# & /@ Range[3]; cons = Flatten@{ Table[(u[j] != #) & /@ vars[[j + 1 ;; -1]], {j, 1, 3 - 1}], 1 vec1 = {1, 2, 3}; vec2 = {1, 2, 3}; NMinimize[ {Total@((vec1[[#]] - Indexed[vec2, u[#]])^2 & /@ R...
Post a Comment
Read more

What is and isn't a valid variable specification for Manipulate?

I have an expression whose terms have arguments (representing subscripts), like this: myExpr = A[0] + V[1,T] I would like to put it inside a Manipulate to see its value as I move around the parameters. (The goal is eventually to plot it wrt one of the variables inside.) However, Mathematica complains when I set V[1,T] as a manipulated variable: Manipulate[Evaluate[myExpr], {A[0], 0, 1}, {V[1, T], 0, 1}] (*Manipulate::vsform: Manipulate argument {V[1,T],0,1} does not have the correct form for a variable specification. >> *) As a workaround, if I get rid of the symbol T inside the argument, it works fine: Manipulate[ Evaluate[myExpr /. T -> 15], {A[0], 0, 1}, {V[1, 15], 0, 1}] Why this behavior? Can anyone point me to the documentation that says what counts as a valid variable? And is there a way to get Manpiulate to accept an expression with a symbolic argument as a variable? Investigations I've done so far: I tried using variableQ from this answer , but it says V[1...
Post a Comment
Read more