def kFoldCrossVal(self, data, fSel, ext, _prune, _info, method, k=5):
acc, md, auc = [], [], []
bef, aft = [], []
chunks = lambda l, n: [l[i:i + n] for i in range(0, len(l), int(n))]
from random import shuffle
rows = data._rows
shuffle(rows)
sqe = chunks(rows, int(len(rows) / k))
if len(sqe) > k:
sqe = sqe[:-2] + [sqe[-2] + sqe[-1]]
for indx in xrange(k):
try:
testRows = sqe.pop(indx)
except:
set_trace()
trainRows = self.flatten([s for s in sqe if not s == testRows])
train, test = clone(data, rows=[
i.cells for i in trainRows]), clone(data, rows=[
i.cells for i in testRows])
train_df = formatData(train)
test_df = formatData(test)
actual = test_df[
test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
_, __, after = self.planner(
train, test, fSel, ext, _prune, _info, method)
bef.extend(before)
aft.extend(after)
md.append((median(before) - median(after)) * 100 / median(before))
auc.append((sum(before) - sum(after)) * 100 / sum(before))
acc.extend(
[(1 - abs(b - a) / a) * 100 for b, a in zip(before, actual)])
sqe.insert(k, testRows)
return acc, auc, md, bef, aft
def mainraw(self, name='Apache', reps=10, fSel=True,
ext=0.5, _prune=False, _info=0.25, method='best'):
data = self.explorer(name)
before, after = [], []
for _ in xrange(reps):
for d in data:
if name == d[0].strip().split('/')[-1]:
train = createTbl([d[0] + '/' + d[1][1]], isBin=False)
test = createTbl([d[0] + '/' + d[1][0]], isBin=False)
train_df = formatData(train)
test_df = formatData(test)
actual = test_df[
test_df.columns[-2]].astype('float32').tolist()
before.append(predictor(train=train_df, test=test_df).rforest())
# set_trace()
newTab = WHAT(
train=[d[0] + '/' + d[1][1]],
test=[d[0] + '/' + d[1][0]],
train_df=train,
bin=True,
test_df=test,
extent=ext,
fSelect=fSel,
far=False,
infoPrune=_info,
method=method,
Prune=_prune).main()
newTab_df = formatData(newTab)
after.append(predictor(train=train_df, test=newTab_df).rforest())
return before, after
def kFoldCrossVal(self, data, fSel, ext, _prune, _info, method, k=5):
acc, md, auc = [], [], []
bef, aft = [], []
chunks = lambda l, n: [l[i:i + n] for i in range(0, len(l), int(n))]
from random import shuffle, sample
rows = data._rows
shuffle(rows)
sqe = chunks(rows, int(len(rows) / k))
if len(sqe) > k:
sqe = sqe[:-2] + [sqe[-2] + sqe[-1]]
for indx in xrange(k):
try:
testRows = sqe.pop(indx)
except:
set_trace()
trainRows = self.flatten([s for s in sqe if not s == testRows])
train, test = clone(data,
rows=[i.cells for i in trainRows]), clone(
data, rows=[i.cells for i in testRows])
train_df = formatData(train)
test_df = formatData(test)
actual = test_df[test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
_, __, after = self.planner(train, test, fSel, ext, _prune, _info,
method)
bef.extend(before)
aft.extend(after)
md.append((median(before) - median(after)) * 100 / median(before))
auc.append((sum(before) - sum(after)) * 100 / sum(before))
acc.extend([(1 - abs(b - a) / a) * 100
for b, a in zip(before, actual)])
sqe.insert(k, testRows)
return acc, auc, md, bef, aft
def planner(self, train, test, fSel, ext, _prune,
_info, name, method='best', justDeltas=False):
train_df = formatData(train)
test_df = formatData(test)
actual = test_df[
test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
# set_trace()
newTab = WHAT(
name=name,
train=None,
test=None,
train_df=train,
bin=True,
test_df=test,
extent=ext,
fSelect=fSel,
far=False,
infoPrune=_info,
method=method,
Prune=_prune).main(justDeltas=justDeltas)
# newTab_df = formatData(newTab)
after = predictor(train=train_df, test=test_df).rforest()
return actual, before, after, newTab
def planner(self, train, test):
train_df = formatData(createTbl(train, _smote=False, isBin=False))
test_df = formatData(createTbl(test, _smote=False, isBin=False))
actual = test_df[test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
# set_trace()
newTab = HOW(train=train, test=test, bin=False).main()
newTab_df = formatData(newTab)
after = predictor(train=train_df, test=newTab_df).rforest()
return newTab
def main(self, name='Apache', reps=20):
out_xtrees = ['xtrees']
out_HOW = ['HOW']
out_cart = ['CART']
out_basln = ['Base']
out_baslnFss = ['Base+FSS']
for _ in xrange(reps):
data = self.explorer(name)
for d in data:
if name == d[0].strip().split('/')[-2]:
# set_trace()
train = [d[0] + d[1][1]]
test = [d[0] + d[1][0]]
# set_trace()
train_df = formatData(createTbl(train, _smote=False, isBin=False))
test_df = formatData(createTbl(test, _smote=False, isBin=False))
actual = test_df[test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
"Apply Different Planners"
xTrees = xtrees(train=train,
test=test,
bin=False,
majority=True).main()
cart = xtrees(train=train,
test=test,
bin=False,
majority=False).main()
how = HOW(name)
baseln = strawman(
train=train,
test=test).main(
config=True)
baselnFss = strawman(
train=train,
test=test,
prune=True).main(config=True)
after = lambda newTab: predictor(
train=train_df,
test=formatData(newTab)).rforest()
frac = lambda aft: sum(aft) / sum(before)
# set_trace()
out_xtrees.append(frac(after(xTrees)))
out_cart.append(frac(after(cart)))
out_HOW.extend(how)
out_basln.append(frac(after(baseln)))
out_baslnFss.append(frac(after(baselnFss)))
return [out_xtrees, out_cart, out_HOW, out_basln, out_baslnFss]
def planner(self, train, test):
train_df = formatData(createTbl(train, _smote=False, isBin=False))
test_df = formatData(createTbl(test, _smote=False, isBin=False))
actual = test_df[
test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
# set_trace()
newTab = HOW(
train=train,
test=test, bin=False).main()
newTab_df = formatData(newTab)
after = predictor(train=train_df, test=newTab_df).rforest()
return newTab
def main(self, name='Apache', reps=20):
out_xtrees = ['xtrees']
out_HOW = ['HOW']
out_cart = ['CART']
out_basln = ['Base']
out_baslnFss = ['Base+FSS']
for _ in xrange(reps):
data = self.explorer(name)
for d in data:
if name == d[0].strip().split('/')[-2]:
# set_trace()
train = [d[0] + d[1][1]]
test = [d[0] + d[1][0]]
# set_trace()
train_df = formatData(
createTbl(train, _smote=False, isBin=False))
test_df = formatData(
createTbl(test, _smote=False, isBin=False))
actual = test_df[test_df.columns[-2]].astype(
'float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
"Apply Different Planners"
xTrees = xtrees(train=train,
test=test,
bin=False,
majority=True).main()
cart = xtrees(train=train,
test=test,
bin=False,
majority=False).main()
how = HOW(name)
baseln = strawman(train=train, test=test).main(config=True)
baselnFss = strawman(train=train, test=test,
prune=True).main(config=True)
after = lambda newTab: predictor(
train=train_df, test=formatData(newTab)).rforest()
frac = lambda aft: sum(aft) / sum(before)
# set_trace()
out_xtrees.append(frac(after(xTrees)))
out_cart.append(frac(after(cart)))
out_HOW.extend(how)
out_basln.append(frac(after(baseln)))
out_baslnFss.append(frac(after(baselnFss)))
return [out_xtrees, out_cart, out_HOW, out_basln, out_baslnFss]
def planner(self, train, test, fSel, ext, _prune, _info, method='best'):
train_df = formatData(train)
test_df = formatData(test)
actual = test_df[test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
# set_trace()
newTab = WHAT(train=None,
test=None,
train_df=train,
bin=True,
test_df=test,
extent=ext,
fSelect=fSel,
far=False,
infoPrune=_info,
method=method,
Prune=_prune).main()
newTab_df = formatData(newTab)
after = predictor(train=train_df, test=newTab_df).rforest()
return actual, before, after
def mainraw(self,
name='Apache',
reps=10,
fSel=True,
ext=0.5,
_prune=False,
_info=0.25,
method='best'):
data = self.explorer(name)
before, after = [], []
for _ in xrange(reps):
for d in data:
if name == d[0].strip().split('/')[-1]:
train = createTbl([d[0] + '/' + d[1][1]], isBin=False)
test = createTbl([d[0] + '/' + d[1][0]], isBin=False)
train_df = formatData(train)
test_df = formatData(test)
actual = test_df[test_df.columns[-2]].astype(
'float32').tolist()
before.append(
predictor(train=train_df, test=test_df).rforest())
# set_trace()
newTab = WHAT(train=[d[0] + '/' + d[1][1]],
test=[d[0] + '/' + d[1][0]],
train_df=train,
bin=True,
test_df=test,
extent=ext,
fSelect=fSel,
far=False,
infoPrune=_info,
method=method,
Prune=_prune).main()
newTab_df = formatData(newTab)
after.append(
predictor(train=train_df, test=newTab_df).rforest())
return before, after
def main(self, name='Apache', reps=20):
rseed(1)
for planner in ['DTREE', 'CD+FS', 'CD', 'BIC']:
out = [planner]
after = lambda newTab: predictor(
train=train_df,
test=formatData(newTab)).rforest()
frac = lambda aft: (1 - sum(aft) / sum(before))
data = self.explorer(name)
for d in data:
if name == d[0].strip().split('/')[-2]:
# set_trace()
train = [d[0] + d[1][1]]
test = [d[0] + d[1][0]]
# set_trace()
train_df = formatData(createTbl(train, _smote=False, isBin=False))
test_df = formatData(createTbl(test, _smote=False, isBin=False))
valid = [
isValid(
new.cells, name=name) for new in createTbl(
test,
_smote=False,
isBin=False)._rows]
actual = test_df[test_df.columns[-2]].astype('float32').tolist()
before = predictor(train=train_df, test=test_df).rforest()
for _ in xrange(reps):
newTab = None # Just so I am sure, there isn't any residue.
"Apply Different Planners"
if planner == 'xtrees':
newTab = xtrees(train=train,
test=test,
bin=False,
majority=True,
name=name).main()
if planner == 'DTREE':
newTab = xtrees(train=train,
test=test,
bin=False,
majority=False,
name=name).main()
valid = [isValid(new.cells, name=name) for new in newTab._rows]
# set_trace()
if planner == 'BIC':
newTab = HOW(name)
valid = [isValid(new.cells, name=name) for new in newTab._rows]
# set_trace()
if planner == 'CD':
newTab = strawman(name=name,
train=train,
test=test).main(mode="config")
valid = [isValid(new.cells, name=name) for new in newTab._rows]
# set_trace()
if planner == 'CD+FS':
newTab = strawman(name=name,
train=train,
test=test,
prune=True).main(mode="config")
valid = [isValid(new.cells, name=name) for new in newTab._rows]
# set_trace()
try:
out.append(frac(after(newTab)))
except:
set_trace()
yield out
