def get_algo(self, train, model=None):
algo_config = self.algo_config
# collect ingoing information
X = train.values
X = X.astype(float)
nominal_ids = set(range(X.shape[1]))
# perform duty
if model is None:
model = Mercs(**algo_config)
tick = time.time()
model.fit(X, nominal_attributes=nominal_ids, **algo_config)
tock = time.time()
fit_time_s = tock - tick
elif isinstance(model, Mercs):
if algo_config["reconfigure_algo"]:
model = self.reconfigure_algo(model, **algo_config)
fit_time_s = model.model_data["ind_time"]
else:
raise ValueError(
"I expect either no model or a Mercs model. Not {}".format(model)
)
return dict(model=model, fit_time_s=fit_time_s)
def test_heart():
# initialize the models
mercs = Mercs(
selection_algorithm="base",
inference_algorithm="base",
prediction_algorithm="it",
max_depth=4,
nb_targets=2,
nb_iterations=1,
n_jobs=1,
verbose=1,
max_steps=8,
)
mercs_mixed = Mercs(
selection_algorithm="base",
inference_algorithm="base",
prediction_algorithm="it",
mixed_algorithm="morfist",
max_depth=4,
nb_targets=2,
nb_iterations=1,
n_jobs=1,
verbose=1,
max_steps=8,
)
# load the data
data = np.loadtxt("./data/heart_failure_clinical_records.csv",
delimiter=",",
skiprows=1)
# split the data into training and testing
x_train, x_test = train_test_split(data, test_size=0.25, random_state=1337)
# ids of the nominal variables
nominal_ids = {1, 3, 5, 9, 10, 12}
# fit the models
mercs.fit(x_train, nominal_attributes=nominal_ids)
mercs_mixed.fit(x_train, nominal_attributes=nominal_ids)
# create the query code for the prediction
q_code = np.zeros(x_train.shape[1], dtype=np.int8)
targets = [2, 12]
q_code[targets] = 1
print(q_code)
# get real values of target variables
y_test = x_test[:, targets]
# predict target values
y_pred = mercs.predict(x_test, q_code=q_code)
y_pred_mixed = mercs_mixed.predict(x_test, q_code=q_code)
scores = compute_scores(y_test, y_pred, [1])
scores_mixed = compute_scores(y_test, y_pred_mixed, [1])
print(scores)
print(scores_mixed)
def get_algo_fit(self, train=None, model=None):
# collect and check ingoing information
algo_config = self.algo_config
assert isinstance(algo_config, dict)
if train is None:
train = self.train
assert isinstance(train, pd.DataFrame)
if model is None:
model = self.model
# actual actions
X = train.values
# X = X.astype(float)
nominal_ids = set(range(X.shape[1]))
if model is None:
model = Mercs(**algo_config)
tick = time.time()
model.fit(X, nominal_attributes=nominal_ids)
tock = time.time()
fit_time_s = tock - tick
elif isinstance(model, Mercs):
fit_time_s = model.model_data["ind_time"]
else:
raise ValueError(
"I expect either no model or a Mercs model. Not {}".format(model)
)
return dict(model=model, fit_time_s=fit_time_s)
def do(self):
# Do the required actions (train, query, storing model...)
# inputs from synth
xlsx_fn = convert_csv_to_xlsx(self.state.filepath)
xl_range_obj = self.get_train_range()
xl_range = None
if xl_range_obj:
xl_range = xl_range_obj.coord
if not xl_range:
print("Range outside of table")
return
wb = load_workbook(xlsx_fn)
empty_rows = get_empty_rows(wb, xl_range_obj)
df, encoders, nominal_ids = extract_data(wb, xl_range, empty_rows)
mod = Mercs(
evaluation_algorithm="dummy",
selection_algorithm="random",
nb_iterations=3,
fraction_missing=[0, 0.2],
)
mod.fit(df.values, nominal_ids=nominal_ids)
q_code = np.zeros(df.shape[1], dtype=int)
empty_cols = get_empty_columns(wb, xl_range_obj)
for c in empty_cols:
q_code[c] = 1
df_pred, _, _ = extract_data(wb, xl_range, [])
y_pred = mod.predict(df_pred.values[:, :-1],
q_code=q_code,
prediction_algorithm="mi")
predictions = []
provenance = ("MERCS", str(uuid4()))
for row in empty_rows:
for col_index, col in enumerate(empty_cols):
value = y_pred[row][col_index].item()
if encoders[col]:
value = encoders[col].inverse_transform([value])[0]
predictions.append(
Prediction(
Coordinate(
col + xl_range_obj.min_col - 1,
row + xl_range_obj.min_row - 1,
),
value,
1,
provenance,
))
return self.state.add_objects(predictions)
def test_init():
# load default dataset and print head
train, test = default_dataset(n_features=3)
# initialise MERCS model
# the nb_targets defines the number of targets to use while fitting the model
# but it is unrelated to the number of targets to use while predicting
clf = Mercs(
selection_algorithm="base",
inference_algorithm="base",
prediction_algorithm="it",
mixed_algorithm="morfist",
max_depth=4,
nb_targets=2,
nb_iterations=2,
n_jobs=1,
verbose=1,
max_steps=8,
)
# ids of the nominal variables
nominal_ids = {train.shape[1]-1}
# fit the model
clf.fit(train, nominal_attributes=nominal_ids)
for idx, m in enumerate(clf.m_list):
msg = """
Model with index: {}
{}
""".format(idx, m.model)
print(msg)
for m_idx, m in enumerate(clf.m_list):
msg = """
Tree with id: {}
has source attributes: {}
has target attributes: {},
and predicts {} attributes
""".format(m_idx, m.desc_ids, m.targ_ids, m.out_kind)
print(msg)
# Query code is [0 0 0 1] where 0 = descriptive and 1 = target
q_code = np.array([0, 0, 0, 1])
print("Query code is:", q_code)
# predict value of query code for test data
y_pred = clf.predict(test, q_code=q_code)
# print the first 10 predictions
print("Predictions:", y_pred[:10])
def mercs():
mercs = Mercs(
selection_algorithm="base",
inference_algorithm="base",
prediction_algorithm="it",
mixed_algorithm="morfist",
max_depth=4,
nb_targets=2,
nb_iterations=2,
n_jobs=1,
verbose=1,
max_steps=8,
)
train, test = default_dataset(n_features=3)
# ids of the nominal variables
nominal_ids = {train.shape[1] - 1}
# fit the model
mercs.fit(train, nominal_attributes=nominal_ids)
return mercs
def test_students():
# initialize the models
mercs = Mercs(
selection_algorithm="base",
inference_algorithm="base",
prediction_algorithm="it",
max_depth=4,
nb_targets=2,
nb_iterations=1,
n_jobs=1,
verbose=1,
max_steps=8,
)
mercs_mixed = Mercs(
selection_algorithm="base",
inference_algorithm="base",
prediction_algorithm="it",
mixed_algorithm="morfist",
max_depth=7,
nb_targets=2,
nb_iterations=1,
n_jobs=1,
verbose=1,
max_steps=8,
)
# load the data
data = np.loadtxt("./data/student-por.csv", delimiter=",", skiprows=1)
# split the data into training and testing
x_train, x_test = train_test_split(data, test_size=0.25)
# ids of the nominal variables
nominal_ids = {
0, 1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 29,
21, 22, 23, 24, 25, 26, 27, 28
}
# fit the models
mercs.fit(x_train, nominal_attributes=nominal_ids)
mercs_mixed.fit(x_train, nominal_attributes=nominal_ids)
# create the query code for the prediction
q_code = np.zeros(x_train.shape[1], dtype=np.int8)
targets = [1, 21, 31]
q_code[targets] = 1
print(q_code)
# get real values of target variables
y_test = x_test[:, targets]
# predict target values
y_pred = mercs.predict(x_test, q_code=q_code)
y_pred_mixed = mercs_mixed.predict(x_test, q_code=q_code)
scores = compute_scores(y_test, y_pred, [0, 1])
scores_mixed = compute_scores(y_test, y_pred_mixed, [0, 1])
print(scores)
print(scores_mixed)