########################## Parameters ################################# loss = 'squarederror' delta = 0.9 grouplist = [(0,3), (3,5), (5, 8), (8,10)] groupweight = None regu = 'grouplasso' reargs = {'lam':0.5} ######################################################################## if loss in ['squarederror', 'huberloss']: fname = 'dataset/regudata' calmod = 'mse' else: fname = 'dataset/clasdata' calmod = 'acc' ######################################################################## # load data set data, target = loadata(fname) # learning and predict ... #args --> (loss, delta, regu, reargs, grouplist, groupweight) LS = GSLM(loss=loss, delta=delta, regu=regu, reargs=reargs, grouplist=grouplist, groupweight=groupweight) LS.fit(data, target) pred = LS.predict(data) res = accmse(target, pred, calmod=calmod) print LS.coeff_ print res
from splearn import lars, lars_path from splearn import cwd_lasso, cwd_elasticnet from splearn import plotpath, loadata from numpy.random import uniform, randn from numpy import dot X, y = loadata('dataset/regudata') caltype = 'lar' path = lars_path(X, y, method=caltype) beta1 = lars(X, y, lam=100) print beta1 beta2 = cwd_lasso(X, y, lam=100) print beta2 beta3 = cwd_elasticnet(X, y, lam1=100, lam2=3.0) print beta3 plotpath(path, pathtype=caltype.upper())
delta = 0.9 grouplist = [(0, 3), (3, 5), (5, 8), (8, 10)] groupweight = None regu = 'grouplasso' reargs = {'lam': 0.5} ######################################################################## if loss in ['squarederror', 'huberloss']: fname = 'dataset/regudata' calmod = 'mse' else: fname = 'dataset/clasdata' calmod = 'acc' ######################################################################## # load data set data, target = loadata(fname) # learning and predict ... #args --> (loss, delta, regu, reargs, grouplist, groupweight) LS = GSLM(loss=loss, delta=delta, regu=regu, reargs=reargs, grouplist=grouplist, groupweight=groupweight) LS.fit(data, target) pred = LS.predict(data) res = accmse(target, pred, calmod=calmod) print LS.coeff_ print res
from splearn import lars, lars_path from splearn import cwd_lasso, cwd_elasticnet from splearn import plotpath, loadata from numpy.random import uniform, randn from numpy import dot X, y = loadata("dataset/regudata") caltype = "lar" path = lars_path(X, y, method=caltype) beta1 = lars(X, y, lam=100) print beta1 beta2 = cwd_lasso(X, y, lam=100) print beta2 beta3 = cwd_elasticnet(X, y, lam1=100, lam2=3.0) print beta3 plotpath(path, pathtype=caltype.upper())