def gbm(X, Y, M=50, gradient=gradient_quad):
N = len(Y)
Model = [np.mean(Y)]
for m in range(M):
print '%d:%f' % (m, RSS(Model, X, Y))
residual = np.zeros(N)
for i in range(N):
residual[i] = gradient(Model, X[i], Y[i])
#print residual
new_tree = RegressionTree.buildtree(X, residual, max_depth=1)
print 'New tree builded.'
#fit(Model, new_tree)
print 'Tree fited.'
Model.append(new_tree)
print '%d:%f' % (m, RSS(Model, X, Y))
return Model
def predict(Model, x, v=0.05):
f = Model[0]
for model in Model[1:]:
res = RegressionTree.predict(x, model)
f += res*v
return f
right = x+(w1+w2)/2
draw.text((x-20,y-10), str(tree.col)+':'+str(tree.value), (0, 0, 0))
draw.line((x, y, left+w1/2, y+100), fill=(255, 0, 0))
draw.line((x, y, right-w2/2, y+100), fill=(255, 0, 0))
drawnode(draw, tree.left_child, left+w1/2, y+100)
drawnode(draw, tree.right_child, right-w2/2, y+100)
else:
txt = '%d' % tree.result
draw.text((x-20, y), txt, (0, 0, 0))
if __name__ == '__main__':
inputs, output, Ttype = prostate.loaddata()
train_data, train_out, test_data, test_out = RegressionTree.cookdata(inputs, output, Ttype)
P = len(train_data[0])
print 'Training...'
Model = gbm(train_data, train_out)
print 'Trained!'
total = 0.
for i in range(len(test_data)):
res = predict(Model, test_data[i])
print res, test_out[i]
total += (res-test_out[i])**2
print total/len(test_out)
print '**************************'
im = relative_importance(Model, P)
for key in im.keys():
print key, im[key]
