def test_min_inquiries(self):
"""Test min inquiries criteria."""
alphabet = ['a', 'b', 'c', 'd']
stopping_criteria = CriteriaEvaluator(
continue_criteria=[MinIterationsCriteria(min_num_inq=2)],
commit_criteria=[
MaxIterationsCriteria(max_num_inq=10),
ProbThresholdCriteria(threshold=0.8)
])
stimuli_agent = NBestStimuliAgent(alphabet=alphabet, len_query=2)
decision_maker = mf.DecisionMaker(stimuli_agent=stimuli_agent,
stopping_evaluator=stopping_criteria,
state='',
alphabet=['a', 'b', 'c', 'd'])
# Initialize with initial (language model) probabilities for each letter.
lm_prior = np.array([0.1, 0.1, 0.1, 0.1])
decision_maker.decide(lm_prior)
likelihood = np.array([0.05, 0.05, 0.09, 0.81])
decision, _arg = decision_maker.decide(likelihood)
self.assertFalse(decision, "Not enough inquiries presented")
likelihood = np.array([0.05, 0.05, 0.09, 0.81])
decision, _arg = decision_maker.decide(likelihood)
self.assertTrue(decision)
self.assertEqual('d', decision_maker.displayed_state)
def test_decision_maker(self):
"""Test default behavior"""
alphabet = ['a', 'b', 'c', 'd']
stopping_criteria = CriteriaEvaluator(
continue_criteria=[MinIterationsCriteria(min_num_inq=1)],
commit_criteria=[
MaxIterationsCriteria(max_num_inq=10),
ProbThresholdCriteria(threshold=0.8)
])
stimuli_agent = NBestStimuliAgent(alphabet=alphabet, len_query=2)
decision_maker = mf.DecisionMaker(stimuli_agent=stimuli_agent,
stopping_evaluator=stopping_criteria,
state='',
alphabet=['a', 'b', 'c', 'd'])
likelihood = np.array([0.2, 0.2, 0.2, 0.4])
decision, _arg = decision_maker.decide(likelihood)
self.assertFalse(decision)
self.assertEqual('.', decision_maker.state)
self.assertEqual('', decision_maker.displayed_state)
likelihood = np.array([0.05, 0.05, 0.1, 0.8])
decision, _arg = decision_maker.decide(likelihood)
self.assertFalse(decision, "Must exceed threshold")
self.assertEqual('..', decision_maker.state)
likelihood = np.array([0.05, 0.05, 0.09, 0.81])
decision, _arg = decision_maker.decide(likelihood)
self.assertTrue(decision)
self.assertEqual('..d', decision_maker.state)
self.assertEqual('d', decision_maker.displayed_state)
def __init__(
self,
state='',
alphabet=list(string.ascii_uppercase) + ['<'] + [SPACE_CHAR],
is_txt_stim=True,
stimuli_timing=[1, .2],
inq_constants=None,
stopping_evaluator=CriteriaEvaluator.default(min_num_inq=2,
max_num_inq=10,
threshold=0.8),
stimuli_agent=RandomStimuliAgent(
alphabet=list(string.ascii_uppercase) + ['<'] + [SPACE_CHAR])):
self.state = state
self.displayed_state = self.form_display_state(state)
self.stimuli_timing = stimuli_timing
# TODO: read from parameters file
self.alphabet = alphabet
self.is_txt_stim = is_txt_stim
self.list_series = [{
'target': None,
'time_spent': 0,
'list_sti': [],
'list_distribution': [],
'decision': None
}]
self.time = 0
self.inquiry_counter = 0
# Stopping Criteria
self.stopping_evaluator = stopping_evaluator
# Stimuli Agent
self.stimuli_agent = stimuli_agent
self.last_selection = ''
# Items shown in every inquiry
self.inq_constants = inq_constants
def setUp(self):
"""Set up decision maker object for testing """
alphabet = list(string.ascii_uppercase) + ['<'] + [SPACE_CHAR]
stopping_criteria = CriteriaEvaluator(
continue_criteria=[MinIterationsCriteria(min_num_inq=1)],
commit_criteria=[
MaxIterationsCriteria(max_num_inq=10),
ProbThresholdCriteria(threshold=0.8)
])
stimuli_agent = NBestStimuliAgent(alphabet=alphabet, len_query=10)
self.decision_maker = DecisionMaker(
stimuli_agent=stimuli_agent,
stopping_evaluator=stopping_criteria,
state='',
alphabet=alphabet,
is_txt_stim=True,
stimuli_timing=[1, .2],
inq_constants=None)
self.evidence_fusion = EvidenceFusion(list_name_evidence=['A', 'B'],
len_dist=2)
def test_max_inquiries(self):
"""Test max inquiries criteria."""
alphabet = ['a', 'b', 'c', 'd']
stopping_criteria = CriteriaEvaluator(
continue_criteria=[MinIterationsCriteria(min_num_inq=1)],
commit_criteria=[
MaxIterationsCriteria(max_num_inq=3),
ProbThresholdCriteria(threshold=0.8)
])
stimuli_agent = NBestStimuliAgent(alphabet=alphabet, len_query=2)
decision_maker = mf.DecisionMaker(stimuli_agent=stimuli_agent,
stopping_evaluator=stopping_criteria,
state='',
alphabet=['a', 'b', 'c', 'd'])
# Initialize with initial (language model) probabilities for each letter.
lm_prior = np.array([0.1, 0.1, 0.1, 0.1])
decision_maker.decide(lm_prior)
likelihood = np.array([0.2, 0.2, 0.4, 0.2])
decision, _arg = decision_maker.decide(likelihood)
self.assertFalse(decision)
likelihood = np.array([0.2, 0.2, 0.4, 0.2])
decision, _arg = decision_maker.decide(likelihood)
self.assertFalse(decision)
likelihood = np.array([0.2, 0.2, 0.4, 0.2])
decision, _arg = decision_maker.decide(likelihood)
self.assertTrue(
decision,
"Should make a decision after max inquiries are reached even if max value is below threshold"
)
self.assertEqual('c', decision_maker.displayed_state)
def test_decision_maker_threshold(self):
"""Threshold should be configurable"""
alphabet = ['a', 'b', 'c', 'd']
stopping_criteria = CriteriaEvaluator(
continue_criteria=[MinIterationsCriteria(min_num_inq=1)],
commit_criteria=[
MaxIterationsCriteria(max_num_inq=10),
ProbThresholdCriteria(threshold=0.5)
])
stimuli_agent = NBestStimuliAgent(alphabet=alphabet, len_query=2)
decision_maker = mf.DecisionMaker(stimuli_agent=stimuli_agent,
stopping_evaluator=stopping_criteria,
state='',
alphabet=['a', 'b', 'c', 'd'])
# Initialize with initial (language model) probabilities for each letter.
lm_prior = np.array([0.1, 0.1, 0.1, 0.1])
decision_maker.decide(lm_prior)
likelihood = np.array([0.2, 0.6, 0.1, 0.1])
decision, _arg = decision_maker.decide(likelihood)
self.assertTrue(decision, "Value should be above configured threshold")
self.assertEqual('b', decision_maker.displayed_state)
def __init__(self,
min_num_inq,
max_num_inq,
signal_model=None,
fs=300,
k=2,
alp=None,
evidence_names=['LM', 'ERP'],
task_list=[('I_LOVE_COOKIES', 'I_LOVE_')],
lmodel=None,
is_txt_stim=True,
device_name='LSL',
device_channels=None,
stimuli_timing=[1, .2],
decision_threshold=0.8,
backspace_prob=0.05,
backspace_always_shown=False,
filter_high=45,
filter_low=2,
filter_order=2,
notch_filter_frequency=60):
self.conjugator = EvidenceFusion(evidence_names, len_dist=len(alp))
inq_constants = []
if backspace_always_shown and BACKSPACE_CHAR in alp:
inq_constants.append(BACKSPACE_CHAR)
# Stimuli Selection Module
stopping_criteria = CriteriaEvaluator(
continue_criteria=[MinIterationsCriteria(min_num_inq)],
commit_criteria=[
MaxIterationsCriteria(max_num_inq),
ProbThresholdCriteria(decision_threshold)
])
# TODO: Parametrize len_query in the future releases!
stimuli_agent = NBestStimuliAgent(alphabet=alp, len_query=10)
self.decision_maker = DecisionMaker(
stimuli_agent=stimuli_agent,
stopping_evaluator=stopping_criteria,
state=task_list[0][1],
alphabet=alp,
is_txt_stim=is_txt_stim,
stimuli_timing=stimuli_timing,
inq_constants=inq_constants)
self.alp = alp
# non-letter target labels include the fixation cross and calibration.
self.nonletters = ['+', 'PLUS', 'calibration_trigger']
self.valid_targets = set(self.alp)
self.signal_model = signal_model
self.sampling_rate = fs
self.downsample_rate = k
self.filter_high = filter_high
self.filter_low = filter_low
self.filter_order = filter_order
self.notch_filter_frequency = notch_filter_frequency
self.mode = 'copy_phrase'
self.task_list = task_list
self.lmodel = lmodel
self.channel_map = analysis_channels(device_channels, device_name)
self.backspace_prob = backspace_prob