def df_to_matrix(df):
""" Converts a CCOBRA dataset into matrix form.
"""
clean = df[["id", "task", "response"]]
usr = list(clean["id"].unique())
matrix = np.zeros((576, len(usr)))
for _, row in clean.iterrows():
usr_idx = usr.index(row["id"])
syl_item = ccobra.Item(usr_idx, "syllogistic", row["task"],
"single-choice", "", 0)
syllog = ccobra.syllogistic.Syllogism(syl_item)
enc_resp = syllog.encode_response(row["response"].split(";"))
syl_idx = ccobra.syllogistic.SYLLOGISMS.index(syllog.encoded_task)
resp_idx = ccobra.syllogistic.RESPONSES.index(enc_resp)
comb_idx = syl_idx * 9 + resp_idx
if matrix[comb_idx, usr_idx] != 0:
print("Tried to write twice to field")
exit()
matrix[comb_idx, usr_idx] = 1
return matrix
def get_train_data_dict(self, ccobra_data):
""" Extracts the training data dict mapping from subject identifiers to their corresponding
list of tasks and responses.
Parameters
----------
ccobra_data : ccobra.CCobraData
Data to convert to the training data representation.
Returns
-------
dict
Dictionary containing the training information (item, response, etc.).
"""
train_data_dict = {}
for id_info, subj_df in ccobra_data.get().groupby('id'):
subj_data = []
for _, row in subj_df.sort_values(['sequence']).iterrows():
train_dict = {}
# Add the Item
train_dict['item'] = ccobra.Item(row['id'], row['domain'],
row['task'],
row['response_type'],
row['choices'],
row['sequence'])
# Convert the response to its list representation
if isinstance(row['response'], str):
if row['response_type'] == 'multiple-choice':
train_dict['response'] = [
y.split(';') for y in
[x.split('/') for x in row['response'].split('|')]
]
else:
train_dict['response'] = [
x.split(';') for x in row['response'].split('/')
]
else:
train_dict['response'] = row['response']
# Add the remaining elements
for key, value in row.iteritems():
if key not in ccobra_data.required_fields:
train_dict[key] = value
subj_data.append(train_dict)
train_data_dict[id_info] = subj_data
return train_data_dict
def persubjectify(dataframe):
""" Convert dataframe to iterable over subjects and their items nested (similar as taken by pre_train)
:param dataframe: pandas dataframe. Required fields: "id", "domain", "task", "response_type", "choices"
:return:
"""
dataset = [[] for i in dataframe["id"].unique()]
# handle the case where ids start at integers other than 0 (e.g. 1)
dataframe["id_normalized"] = dataframe["id"] - dataframe["id"].min()
for i, row in dataframe.iterrows():
subject_id = row["id_normalized"]
item = ccobra.Item(identifier=subject_id,
domain=row["domain"],
task=row["task"],
resp_type=row["response_type"],
choices=row["choices"])
response = [row["response"].split(";")]
dataset[subject_id].append({"item": item, "response": response})
return dataset
def predict(self, item, **kwargs):
"""
Generate random response prediction.
"""
model = np.zeros((6, 3), dtype=bool)
syl = ccobra.syllogistic_generalized.GeneralizedSyllogism(item)
if syl.figure == 2:
prs = item.task_str.split("/")
new_task = prs[1] + "/" + prs[0]
new_item = ccobra.Item(item.identifier, item.domain, new_task,
item.response_type, item.choices_str,
item.sequence_number)
syl = ccobra.syllogistic_generalized.GeneralizedSyllogism(new_item)
instances = bidict({syl.A: 0, syl.B: 1, syl.C: 2})
print(instances)
# Fill first premise
q1, i0, i1 = syl.p1
model = self.fill_first_premise(model, q1, instances[i0], instances[i1])
# Fill model with second premise
q2, i2, i3 = syl.p2
model = self.fill_second_premise(model, q2, 1, 2, 0)
print(syl.task)
print(model.astype(int))
for quantifier in self.axioms:
print("Conclusion", instances[i0], instances[i1])
if self.axioms[quantifier](model, 0, 2):
print('Prediction', quantifier, syl.A, syl.C)
return [[quantifier, syl.A, syl.C]]
print('No conclusion possible')
return [['NVC']]
def syllogism_to_item(syllogism):
"""
>>> syllogism_to_item("AA1").task
[['All', 'x', 'y'], ['All', 'y', 'z']]
>>> syllogism_to_item("EO4").task
[['No', 'y', 'x'], ['Some not', 'y', 'z']]
>>> syllogism_to_item("II2").task
[['Some', 'y', 'x'], ['Some', 'z', 'y']]
"""
q1 = {"A": "All", "I": "Some", "E": "No", "O": "Some not"}[syllogism[0]]
q2 = {"A": "All", "I": "Some", "E": "No", "O": "Some not"}[syllogism[1]]
to = {
"1": ["x;y", "y;z"],
"2": ["y;x", "z;y"],
"3": ["x;y", "z;y"],
"4": ["y;x", "y;z"]
}[syllogism[2]]
# task = e.g. "All;x;y/All;y;z",
task = q1 + ";" + to[0] + "/" + q2 + ";" + to[1]
return ccobra.Item(0, "syllogistic", task, "single-choice",
GENERIC_CHOICES)
"Some;y;x/Some not;z;y", "Some;x;y/Some not;z;y", "Some;y;x/Some not;y;z",
"No;x;y/All;y;z", "No;y;x/All;z;y", "No;x;y/All;z;y", "No;y;x/All;y;z",
"No;x;y/Some;y;z", "No;y;x/Some;z;y", "No;x;y/Some;z;y", "No;y;x/Some;y;z",
"No;x;y/No;y;z", "No;y;x/No;z;y", "No;x;y/No;z;y", "No;y;x/No;y;z",
"No;x;y/Some not;y;z", "No;y;x/Some not;z;y", "No;x;y/Some not;z;y",
"No;y;x/Some not;y;z", "Some not;x;y/All;y;z", "Some not;y;x/All;z;y",
"Some not;x;y/All;z;y", "Some not;y;x/All;y;z", "Some not;x;y/Some;y;z",
"Some not;y;x/Some;z;y", "Some not;x;y/Some;z;y", "Some not;y;x/Some;y;z",
"Some not;x;y/No;y;z", "Some not;y;x/No;z;y", "Some not;x;y/No;z;y",
"Some not;y;x/No;y;z", "Some not;x;y/Some not;y;z",
"Some not;y;x/Some not;z;y", "Some not;x;y/Some not;z;y",
"Some not;y;x/Some not;y;z"
]
GENERIC_CHOICES = "All/x/z|All/z/x|Some/x/z|Some/z/x|Some not/x/z|Some not/z/x|No/x/z|No/z/x|NVC"
GENERIC_ITEMS = [
ccobra.Item(0, "syllogistic", task, "single-choice", GENERIC_CHOICES)
for task in GENERIC_TASKS
]
# ---- SYLLOGISTIC FUNCTIONS ---- #
def add_implicatures(conclusions, existential=True, gricean=True):
"""
existential: A -> I, E -> O
gricean: I -> O, O -> I
>>> add_implicatures(["Aac", "Aca"], gricean=False)
['Aac', 'Aca', 'Iac', 'Ica']
>>> add_implicatures(["Aac", "Eac", "NVC"], gricean=False)
['Aac', 'Eac', 'NVC', 'Iac', 'Oac']
"Few;A;B/Few;B;C", "Few;A;B/Most;B;C", "Few;A;B/No;B;C",
"Few;A;B/Some not;B;C", "Few;A;B/Some;B;C", "Most;A;B/All;B;C",
"Most;A;B/Few;B;C", "Most;A;B/Most;B;C", "Most;A;B/No;B;C",
"Most;A;B/Some not;B;C", "Most;A;B/Some;B;C", "No;A;B/Few;B;C",
"No;A;B/Most;B;C", "Some not;A;B/Few;B;C", "Some not;A;B/Most;B;C",
"Some;A;B/Few;B;C", "Some;A;B/Most;B;C"
# "All;A;B/Few;B;C"
# ,"All;A;B/Few;B;C"
# "Some;fencers;engineers/Some;campers;fencers"
# ,"Most not;waiters;poets/Some;cashiers;waiters"
]
choices = [
"All;A;C|Some;A;C|No;A;C|Some not;A;C|Most;A;C|Few;A;C"
# "All;C;A|Some;C;A|No;C;A|Some not;C;A|Most;C;A|Few;C;A"
# "All;engineers;campers|All;campers;engineers|Some;engineers;campers
# |Some;campers;engineers|No;engineers;campers|No;campers;engineers|Some not;engineers;campers|Some not;campers;engineers|Most;engineers;campers|Most;campers;engineers|Most not;engineers;campers|Most not;campers;engineers|NVC"
# ,"All;poets;cashiers|All;cashiers;poets|Some;poets;cashiers|Some
# ;cashiers;poets|No;poets;cashiers|No;cashiers;poets|Some not;poets;cashiers|Some not;cashiers;poets|Most;poets;cashiers|Most;cashiers;poets|Most not;poets;cashiers|Most not;cashiers;poets|NVC"
]
# single_choice = "All;A;C|Some;A;C|No;A;C|Some not;A;C|Most;A;C|Few;A;C"
single_choice = "All;C;A|Some;C;A|No;C;A|Some not;C;A|Most;C;A|Few;C;A"
idx = 0
for task in tasks:
test_item = ccobra.Item(identifier, domain, task, resp_type,
single_choice, sequence_number)
result = pm.predict(test_item)
print("--------------------------------------")
idx += 1
return [['NVC']]
if __name__ == "__main__":
# execute only if run as a script
pm = PreferredModel()
identifier = "TEST"
sequence_number = 1
domain = "syllogistic-generalized"
resp_type = "single-choice"
tasks = [
"All;A;B/Few;B;C"
,"Some;fencers;engineers/Some;campers;fencers"
,"Most not;waiters;poets/Some;cashiers;waiters"
]
choices = [
"All;A;C|Some;A;C|No;A;C|Some not;A;C|Most;A;C|Few;A;C"
,"All;engineers;campers|All;campers;engineers|Some;engineers;campers|Some;campers;engineers|No;engineers;campers|No;campers;engineers|Some not;engineers;campers|Some not;campers;engineers|Most;engineers;campers|Most;campers;engineers|Most not;engineers;campers|Most not;campers;engineers|NVC"
,"All;poets;cashiers|All;cashiers;poets|Some;poets;cashiers|Some;cashiers;poets|No;poets;cashiers|No;cashiers;poets|Some not;poets;cashiers|Some not;cashiers;poets|Most;poets;cashiers|Most;cashiers;poets|Most not;poets;cashiers|Most not;cashiers;poets|NVC"
]
idx = 0;
for task in tasks:
test_item = ccobra.Item(identifier, domain, task, resp_type,
choices[idx], sequence_number)
result = pm.predict(test_item)
print("--------------------------------------")
idx += 1