def divide(self, lo, hi, rank):
"Find a split between lo and hi, then recurse on each split."
xleft = Num(key=self.x)
yleft = self.ctype(key=self.y)
xright = Num(self.lst[lo:hi], key=self.x)
yright = self.ctype(self.lst[lo:hi], key=self.y)
xb4 = deepcopy(xright)
yb4 = deepcopy(yright)
best = yb4.variety()
cut = None
for j in range(lo, hi):
xleft + self.lst[j]
yleft + self.lst[j]
xright - self.lst[j]
yright - self.lst[j]
if xleft.n >= self.step:
if xright.n >= self.step:
now = self.x(self.lst[j - 1])
after = self.x(self.lst[j])
if now == after: continue
if abs(xright.mu - xleft.mu) >= self.epsilon:
if after - self.start >= self.epsilon:
if self.stop - now >= self.epsilon:
xpect = yleft.xpect(yright)
if xpect * THE.div.trivial < best:
best, cut = xpect, j
if cut:
rank = self.divide(lo, cut, rank) + 1
rank = self.divide(cut, hi, rank)
else:
self.gain += xb4.n * xb4.variety()
xb4.rank = rank
self.ranges += [(xb4, yb4)]
return rank
def createXYList(i, lst, yis, y_index):
# print("list!!", lst)
# print("List before: ",lst)
if yis == "Num":
y_lst = Num()
else:
y_lst = Sym()
x_lst = []
lst = sorted(lst, key=lambda x: x.cells[y_index])
for column in range(len(lst[0].cells) -
1): #last column excluded for goal
if not column == y_index:
temp = Num()
for row in lst:
temp.add(row.cells[column])
# print("temp: ", temp)
x_lst.append(temp)
for row in lst:
y_lst.add(row.cells[y_index])
# print(x_lst)
return x_lst, y_lst
def createXYList(i, lst, yis):
lst = sorted(lst, key=lambda x: x[1])
x_lst = Num()
if yis == "Num":
y_lst = Num()
else:
y_lst = Sym()
for i in lst:
x_lst.add(i[0])
y_lst.add(i[1])
return x_lst, y_lst
def tree(i, lst, y, yis, lvl=0):
if len(lst) >= THE.tree.minObs * 2:
# find the best column
lo, cut, col = -10**32, None, None
for col1 in i.cols.indep:
x = lambda row: row.cells[col1.pos]
d = Div2(lst, x=x, y=y, yis=yis)
cut1, lo1 = d.finalcutlow()
# print(cut1, lo1)
if cut1:
if lo1 < lo:
cut, lo, col = cut1, lo1, col1
print("updated: ", cut, lo, col)
# if a cut exists
if cut:
# split data on best col, call i.tree on each split
x = lambda row: row.cells[col.pos]
return [
o(lo=lo,
hi=hi,
n=len(kids),
txt=col.txt,
kids=i.tree(kids, y, yis, lvl + 1))
for lo, hi, kids in col.split(lst, x, cut)
]
if yis == "Num":
return Num(lst, key=y)
else:
return Sym(lst, key=y)
class Div2_3(Pretty):
"""
Recursively divide a list of numns by finding splits
that minimizing the expected value of the standard
deviation (after the splits).
"""
def __init__(self, lst, x=first, y=last, yis=Num):
self.ctype = yis
self.x = x
self.y = y
self.lst = ordered(lst, key=x)
self.xtype = Num(self.lst, key=x)
self.ytype = self.ctype(self.lst, key=y)
self.gain = 0 # where we will be, once done
#i.x = x # how to get values from 'lst' items
self.step = int(len(
self.lst)**THE.div.min) # each split need >= 'step' items
self.stop = x(last(self.lst)) # top list value
self.start = x(first(self.lst)) # bottom list value
self.ranges = [] # the generted ranges
self.epsilon = self.xtype.sd(
) * THE.div.cohen # bins must be seperated >= epsilon
self.divide(1, len(self.lst), 1)
self.gain /= len(self.lst)
#TODO: check the argument passing thing
def divide(self, lo, hi, rank):
"Find a split between lo and hi, then recurse on each split."
xleft = Num(key=self.x)
yleft = self.ctype(key=self.y)
xright = Num(self.lst[lo:hi], key=self.x)
yright = self.ctype(self.lst[lo:hi], key=self.y)
xb4 = deepcopy(xright)
yb4 = deepcopy(yright)
best = yb4.variety()
cut = None
for j in range(lo, hi):
xleft + self.lst[j]
yleft + self.lst[j]
xright - self.lst[j]
yright - self.lst[j]
if xleft.n >= self.step:
if xright.n >= self.step:
now = self.x(self.lst[j - 1])
after = self.x(self.lst[j])
if now == after: continue
if abs(xright.mu - xleft.mu) >= self.epsilon:
if after - self.start >= self.epsilon:
if self.stop - now >= self.epsilon:
xpect = yleft.xpect(yright)
if xpect * THE.div.trivial < best:
best, cut = xpect, j
if cut:
rank = self.divide(lo, cut, rank) + 1
rank = self.divide(cut, hi, rank)
else:
self.gain += xb4.n * xb4.variety()
xb4.rank = rank
self.ranges += [(xb4, yb4)]
return rank
def __init__(self, lst, x=first, y=last, yis=Num):
self.ctype = yis
self.x = x
self.y = y
self.lst = ordered(lst, key=x)
self.xtype = Num(self.lst, key=x)
self.ytype = self.ctype(self.lst, key=y)
self.gain = 0 # where we will be, once done
#i.x = x # how to get values from 'lst' items
self.step = int(len(
self.lst)**THE.div.min) # each split need >= 'step' items
self.stop = x(last(self.lst)) # top list value
self.start = x(first(self.lst)) # bottom list value
self.ranges = [] # the generted ranges
self.epsilon = self.xtype.sd(
) * THE.div.cohen # bins must be seperated >= epsilon
self.divide(1, len(self.lst), 1)
self.gain /= len(self.lst)
def numSplit(i, lst):
newNumber = Num()
if lst:
for x in lst:
newNumber.add(x)
return newNumber