27 lines
1.2 KiB
Python
27 lines
1.2 KiB
Python
from girth import ability_mle
|
|
|
|
from lib.irt.models.base import *
|
|
|
|
class Rasch(Base):
|
|
|
|
def result(self):
|
|
return (1 / (1 + self.e(-1 * (self.theta - self.b_param))))
|
|
|
|
@classmethod
|
|
def ability_estimate(self, items) -> float:
|
|
# responses are mapped into a matrix, where each row and item
|
|
# and each column is an exam form result
|
|
# we'll likely have to change this to something more robust
|
|
# when we get into more complex response types
|
|
responses = np.array([[int(item.response)] for item in items])
|
|
# the difficulty (b param) for each item is in an ordered list
|
|
difficulty = np.array([item.b_param for item in items])
|
|
# the package currently utilizes a fixed a param (discrimination)
|
|
discrimination = np.linspace(1, 1, len(difficulty))
|
|
# there are many methodologies to calculate ability from a data set of responses
|
|
# this is what our client currently uses but we should expand this to allow for
|
|
# switching between methodologies when needed
|
|
# it also currrently only does a single ability estimation
|
|
# at some point we can also accommodate batch ability estimates if need be
|
|
return ability_mle(responses, difficulty, discrimination).tolist()[0]
|