def choose_card(db_conn, user, unit):
"""
Given a user and a unit, choose an appropriate card.
Return a card instance.
"""
# TODO-3 simplify this method
unit_id = unit['entity_id']
query = (Card.start_accepted_query().filter({
'unit_id': unit_id
}).sample(10)) # TODO-2 index
# TODO-3 does this belong as a model method?
# TODO-2 is the sample value decent?
# TODO-2 has the learner seen this card recently?
cards = [Card(d) for d in query.run(db_conn)]
if not len(cards):
return None
previous_response = get_latest_response(user['id'], unit_id, db_conn)
if previous_response:
learned = previous_response['learned']
# Don't allow the previous card as the next card
cards = [
card for card in cards
if card['entity_id'] != previous_response['card_id']
]
else:
learned = init_learned
shuffle(cards)
assessment, nonassessment = partition(cards, lambda c: c.has_assessment())
choose_assessment = random() < p_assessment_map[floor(learned * 10)]
if choose_assessment:
if not len(assessment):
return nonassessment[0]
for card in assessment:
params = get_card_parameters({'entity_id': card['entity_id']},
db_conn)
if params:
values = get_card_parameters_values(params)
guess = values['guess']
slip = values['slip']
correct = calculate_correct(guess, slip, learned)
if 0.25 < correct < 0.75:
return card
else:
return card
return assessment[0]
if len(nonassessment):
return nonassessment[0]
if len(assessment):
return assessment[0]
return None
def test_calculate_correct():
assert calculate_correct(guess=0, slip=0, learned=0.5) == 0.5
assert calculate_correct(guess=1, slip=1, learned=0.5) == 0.5
assert calculate_correct(guess=0, slip=1, learned=0.5) == 0
assert calculate_correct(guess=1, slip=0, learned=0.5) == 1
assert calculate_correct(guess=0.25, slip=0.05, learned=0.9) == 0.88
assert calculate_correct(guess=0.25, slip=0.05, learned=0.5) == 0.6
assert calculate_correct(guess=0.25, slip=0.05, learned=0.1) == 0.32
def likelihood(self, data, hypothesis):
"""
Given new data and one of the guess hypotheses,
update the probability of that hypothesis.
"""
score, learned, slip = \
data['score'], data['learned'], data['slip']
return (score * calculate_correct(hypothesis, slip, learned) +
(1 - score) * calculate_incorrect(hypothesis, slip, learned))
def get_slip_pmf_likelihood(data, hypothesis):
"""
Given new data and one of the slip hypotheses,
update the probability of that hypothesis.
"""
score, learned, guess = \
data['score'], data['learned'], data['guess']
return (score * calculate_correct(guess, hypothesis, learned) +
(1 - score) * calculate_incorrect(guess, hypothesis, learned))
def choose_scored_card(scored, all_params, learned):
for card in scored:
params = all_params[card['entity_id']]
if params:
guess = params['guess']
slip = params['slip']
correct = calculate_correct(guess, slip, learned)
if 0.25 < correct < 0.75:
return card
return scored[0]
def likelihood(self, data, hypothesis):
"""
Given new data and one of the slip hypotheses,
update the probability of that hypothesis.
"""
score, learned, guess = \
data['score'], data['learned'], data['guess']
return (score
* calculate_correct(guess, hypothesis, learned)
+ (1 - score)
* calculate_incorrect(guess, hypothesis, learned))
def slip_normal(score, guess, slip, learned):
if score == 1:
return calculate_correct(guess, slip, learned)
if score == 0:
return calculate_incorrect(guess, slip, learned)
def slip_likelihood(score, guess, slip, learned):
if score == 1:
return calculate_correct(guess, 1, learned)
if score == 0:
return calculate_incorrect(guess, 1, learned)
def slip_normal(score, guess, slip, learned):
if score == 1:
return calculate_correct(guess, slip, learned)
if score == 0:
return calculate_incorrect(guess, slip, learned)
def slip_likelihood(score, guess, slip, learned):
if score == 1:
return calculate_correct(guess, 1, learned)
if score == 0:
return calculate_incorrect(guess, 1, learned)
