def __init__(self, model_type, codes, feature_names=None, alphas=None, **kwargs):
self.model_type = model_type
self.models = {}
for i, code in enumerate(codes):
model = SparseModel(model_type, feature_names, **kwargs)
if alphas is not None:
model.set_alpha(alphas[i])
self.models[code] = model
def fit(self, X, all_y):
converted = False
if len(all_y.shape) == 1:
converted = True
else:
n, n_codes = all_y.shape
if n_codes == 1:
converted = True
if converted:
SparseModel.fit(self, X, all_y)
else:
powerset_y = self.binary_vectors_to_powerset_list(all_y)
SparseModel.fit(self, X, powerset_y)
def __init__(self, model_type, codes, feature_names=None, alphas=None, **kwargs):
# generate a random order over models
self.order = list(codes)[:]
random.shuffle(self.order)
self.model_type = model_type
self.models = {}
# create models in the pre-determined order, adding one feature each time for the output of the previous model
for i, code in enumerate(self.order):
model = SparseModel(model_type, feature_names[:], **kwargs)
if alphas is not None:
model.set_alpha(alphas[i])
self.models[code] = model
feature_names.append(code)
def tune_by_cv(self, X, all_y, alpha_values, td_splits, n_dev_folds, reuser=None, verbose=1):
y = self.binary_vectors_to_powerset_list(all_y.as_matrix())
alphas = pd.DataFrame(np.zeros([1, 1]), index=['alpha'], columns=['powerset_label'])
valid_f1_summary, best_alpha = SparseModel.tune_by_cv(self, X, y, alpha_values, td_splits, n_dev_folds,
reuser, verbose)
alphas['alpha', 'powerset_label'] = best_alpha
return alphas
def predict(self, X, index=None, codes=None):
n, p = X.shape
powerset_predictions = SparseModel.predict(self, X)
example_str = self.powerset_labels[0]
example = [int(i) for i in example_str[1:-1].split(',')]
n_codes = len(example)
predictions = np.zeros([n, n_codes])
for i in range(n):
predictions[i, :] = self.powerset_index_to_binary_vec(powerset_predictions[i])
if index is not None:
predictions = pd.DataFrame(predictions, index=index, columns=codes)
return predictions
def __init__(self, model_type, codes, **kwargs):
self.codes = codes
code = ['powerset_label']
SparseModel.__init__(self, model_type, code, **kwargs)
