Prepare Text for Machine Learning$$

If we are to create a classifier for text, we'll first need to think about the format of our data. Take a look at the files girls.train and boys.train. For example with the unix command:

$$

cat girls.train

$...$

Addisyn

Danika

Emilee

Aurora

Julianna

Sophia

Kaylyn

Litzy

Hadassah

This file contains names that are more or less commonly used for girls. The problem with the current data in this file is that the names are in plain text, which is not a format our machine learning algorithm can work with effectively. You need to transform these plain text names into some vector format, where each name becomes a vector that represents a point in some high dimensional input space.

That is exactly what the following Python function name$2$features does, by arbitrarily chunking and hashing different string extractions from each baby name inputted, thus transforming the string into a quantitative feature vector:

$$

def hashfeatures$($baby$,$ d$,$ FIX, debug$=$False$)$:

$"""$

$$ Input:

$$ baby : a string representing the baby's name to be hashed

$$ d: the number of dimensions to be in the feature vector

$$ FIX: the number of chunks to extract and hash from each string

$$ debug: a bool for printing debug values $($default False$)$

$$ Output:

$$ v: a feature vector representing the input string

$"""$

$$ v $=$ np$.$zeros$($d$)$

$$ for m in range$(1,$ FIX$+1)$:

$$ prefix $=$ baby$[$:m$]+">"$

$$ P $=$ hash$($prefix$)\%$ d

$$ v$[$P$]=1$

$$ suffix $="<"+$ baby$[-$m:$]$

$$ S $=$ hash$($suffix$)\%$ d

$$ v$[$S$]=1$

$$ if debug:

$$ print$($f$"$Split $\{$m$\}/\{$FIX$\}$:$\backslash$t$(\{$prefix$\},\{$suffix$\}),\backslash$t$1$s at indices $[\{$P$\},\{$S$\}]")$

$$ if debug:

$$ print$($f$"$Feature vector for $\{$baby$\}$:

$\{$v$.$astype$($int$)\}$

$")$

$$ return v