Python's array has more type than python it self, for example
python only has float while array support 'f' (float) and 'd' (double)
Switching to array.typecode helps spark make a better inference

For example, for the code:

from pyspark.sql.types import _infer_type
from array import array
a = array('f',[1,2,3,4,5,6])
_infer_type(a)

We will get ArrayType(DoubleType,true) before change,
but ArrayType(FloatType,true) after change