Well my question is a general question. I tried to find some relevant information before posting my question here, but no success!.
I am working on
20 newsgroup data set, which as you may know it contains
20 category and in each category there are
for feeding this data set into an machine learning model, we may have several options(correct me if Im wrong please):
matrix in which
number of all docsand the
number of features(for example we decide to extract 1000 features overall). so it will be
doc - word matrices.
(2000,1000). this case in topic modeling approaches
when I use
Scikit-learnlibrary it will give me dimension as
number of samplesand
1000is number of
extracted featuresas column. so probably I can call it
word feature matrix. this case in classification.
we may have a matrix
(11314,20). in this case
number of samples or unique words, and
number of classes(20 news group has 20 classes). I have seen this format when we want to calculate Information Gain which makes sense as we want information gain for each sample in the corpus correspond to each label.
So lets talk about my confusion part:
does it make any difference if I feed my machine learning model with any of these matrices?
I have seen the case 1 and 2 to be used in the machine learning model, does matrix 3
(number of samples, number of classes)does not make sense to be fed to machine learning model?
(when we are using
Information gain the matrix I am using is
(number of samples, number of classes), when I want to change to machine learning model it will change to either
(documents, features) or
(number of samples, features), I am confused when should I use which one).