def argmin(c, lo, hi):
cut = None
xl, yl = Num(), Num(
) # left split for both features and label cols
xr, yr = Num(), Num(
) # right split for both features and label cols
# push everything in the right
for i in range(lo, hi):
xr.numInc(rows[i][c])
yr.numInc(rows[i][goal])
best_x = xr.sd # currently all data is in right so best is sd on right
best_y = yr.sd # currently all data is in right so best is sd on right
mu = yr.mu
# print(best)
# push to the left one by one and keep track of best
if hi - lo > 2 * enough:
for i in range(lo, hi):
x = rows[i][c]
y = rows[i][goal]
xl.numInc(x)
yl.numInc(y)
xr.numDec(x)
yr.numDec(y)
if xl.n >= enough and xr.n >= enough:
tmp_x = xl.numXpect(xr) * 1.05
tmp_y = yl.numXpect(yr) * 1.05
# print(tmp, x)
try:
if tmp_x < best_x:
if tmp_y < best_y:
cut, best_x, best_y = i, tmp_x, tmp_y
# print(tmp_x, tmp_y, best_x, best_y)
except:
print(tmp_x, tmp_y)
return cut, mu
def argmin(c, lo, hi):
cut = None
if hi - lo > 2 * enough:
l = Num() # left split
r = Num() # right split
# push everything in the right
for i in range(lo, hi):
r.numInc(rows[i][c])
best = r.sd # currently all data is in right so best is sd on right
# print(best)
# push to the left one by one and keep track of best
for i in range(lo, hi):
x = rows[i][c]
l.numInc(x)
r.numDec(x)
if l.n >= enough and r.n >= enough:
tmp = Num.numXpect(l, r) * 1.05
# print(tmp, x)
if tmp < best:
cut, best = i, tmp
# print(tmp, best)
return cut