def _RF():
"Test RF"
dir = "../Data"
one, two = explore(dir)
# Training data
train_DF = createTbl([one[0][0]])
# Test data
test_df = createTbl([one[0][1]])
actual = Bugs(test_df)
preds = rforest(train_DF, test_df, mss=6, msl=8, max_feat=4, n_est=5756, smoteit=False)
print _Abcd(before=actual, after=preds, show=False)[-1]
def _logit():
"Test LOGIT"
dir = './Data'
one, two = explore(dir)
# Training data
train_DF = createTbl(one[0])
# Test data
test_df = createTbl(two[0])
actual = Bugs(test_df)
preds = logit(train_DF, test_df)
set_trace()
_Abcd(train=actual, test=preds, verbose=True)
def _adaboost():
"Test AdaBoost"
dir = "./Data"
one, two = explore(dir)
# Training data
train_DF = createTbl(one[0])
# Test data
test_df = createTbl(two[0])
actual = Bugs(test_df)
preds = adaboost(train_DF, test_df)
set_trace()
_Abcd(train=actual, test=preds, verbose=True)
def _logit(): "Test LOGIT" dir = './Data' one, two = explore(dir) # Training data train_DF = createTbl(one[0]) # Test data test_df = createTbl(two[0]) actual = Bugs(test_df) preds = logit(train_DF, test_df) set_trace() _Abcd(train=actual, test=preds, verbose=True)
def _where2pred(): "Test where2" dir = '../Data' one, two = explore(dir) # Training data train_DF = createTbl(one[0]) # Test data test_df = createTbl(two[0]) actual = Bugs(test_df) preds = where2prd(train_DF, test_df) for a, b in zip(actual, preds): print a, b set_trace() print _Abcd(before = actual, after = preds, show = False)
def _RF():
"Test RF"
dir = '../Data'
one, two = explore(dir)
# Training data
train_DF = createTbl([one[0][0]])
# Test data
test_df = createTbl([one[0][1]])
actual = Bugs(test_df)
preds = rforest(train_DF, test_df, mss=6, msl=8,
max_feat=4, n_est=5756,
smoteit=False)
print _Abcd(before=actual, after=preds, show=False)[-1]
def _where2pred():
"Test where2"
dir = '../Data'
one, two = explore(dir)
# Training data
train_DF = createTbl(one[0])
# Test data
test_df = createTbl(two[0])
actual = Bugs(test_df)
preds = where2prd(train_DF, test_df)
for a, b in zip(actual, preds):
print a, b
set_trace()
print _Abcd(before=actual, after=preds, show=False)
def go(self):
for _ in xrange(self.reps):
predRows = []
train_DF = createTbl(self.train[self._n], isBin=True)
test_df = createTbl(self.test[self._n], isBin=True)
actual = np.array(Bugs(test_df))
before = self.pred(train_DF, test_df,
tunings=self.tunedParams,
smoteit=True)
predRows = [row.cells for predicted
, row in zip(before, test_df._rows) if predicted > 0]
predTest = clone(test_df, rows=predRows)
newTab = xtrees(train=self.train[self._n]
, test_DF=predTest, bin=False).main()
after = self.pred(train_DF, newTab,
tunings=self.tunedParams,
smoteit=True)
self.out_pred.append(_Abcd(before=actual, after=before))
# set_trace()
delta = cliffs(lst2=Bugs(predTest), lst1=after).delta()
frac = sum([0 if a < 1 else 1 for a in after]) / \
sum([0 if b < 1 else 1 for b in before])
self.out.append(frac)
print(self.out)
def depen(self, rows):
mod = rforest(
self.train,
self.test,
tunings=rows # n_est, max_feat, mss, msl
,
smoteit=True)
g = _Abcd(before=Bugs(self.test), after=mod, show=False)[-1]
return g
def depen(self, row):
# My where2pred() takes data in string format. Ex: '../Data/ant/ant-1.6.csv'
self.where = defaults().update(minSize = row[4]
, depthMin = int(row[5])
, depthMax = int(row[6])
, prune = row[7]>0.5)
self.tree.infoPrune = row[1]
self.tree.m = int(row[2])
self.tree.n = int(row[3])
self.tree.prune = row[8]>0.5
actual = Bugs(createTbl([self.test], isBin = True))
preds = where2prd(self.train, [self.test], tunings = [self.where, self.tree], thresh = row[0])
return _Abcd(before = actual, after = preds, show = False)[-1]
def _where2pred(): "Test where2" dir = '../Data' one, two = explore(dir) # set_trace() # Training data train = one[0][:-1] # Test data test = [one[0][-1]] actual = Bugs(createTbl(test, isBin=True)) preds = where2prd(train, test) # for a, b in zip(actual, preds): print a, b # set_trace() return _Abcd(before=actual, after=preds, show=False)[-1]
def _where2pred():
"Test where2"
dir = '../Data'
one, two = explore(dir)
# set_trace()
# Training data
train = one[0][:-1]
# Test data
test = [one[0][-1]]
actual = Bugs(createTbl(test, isBin=True))
preds = where2prd(train, test)
# for a, b in zip(actual, preds): print a, b
# set_trace()
return _Abcd(before=actual, after=preds, show=False)[-1]
def depen(self, row):
# My where2pred() takes data in string format. Ex:
# '../Data/ant/ant-1.6.csv'
self.where = defaults().update(
minSize=row[4], depthMin=int(
row[5]), depthMax=int(
row[6]), prune=row[7] > 0.5)
self.tree.infoPrune = row[1]
self.tree.m = int(row[2])
self.tree.n = int(row[3])
self.tree.prune = row[8] > 0.5
actual = Bugs(createTbl([self.test], isBin=True))
preds = where2prd(
self.train, [
self.test], tunings=[
self.where, self.tree], thresh=row[0])
return _Abcd(before=actual, after=preds, show=False)[-1]
def go(self):
for _ in xrange(self.reps):
predRows = []
train_DF = createTbl(self.train[self._n][-2:], isBin=True)
test_df = createTbl(self.test[self._n], isBin=True)
actual = Bugs(test_df)
before = self.pred(train_DF, test_df,
tunings=self.tunedParams,
smoteit=True)
for predicted, row in zip(before, test_df._rows):
tmp = row.cells
tmp[-2] = predicted
if predicted > 0:
predRows.append(tmp)
predTest = clone(test_df, rows=predRows)
if predRows:
newTab = treatments2(
train=self.train[self._n][-2:],
test=self.test[self._n],
test_df=predTest,
extent=self.extent,
far=False,
smote=True,
resample=False,
infoPrune=self.infoPrune,
Prune=self.Prune).main()
else:
newTab = treatments2(
train=self.train[
self._n][-2:],
test=self.test[
self._n],
far=False,
smote=True,
resample=False,
extent=self.extent,
infoPrune=self.infoPrune,
Prune=self.Prune).main()
after = self.pred(train_DF, newTab,
tunings=self.tunedParams,
smoteit=True)
self.out_pred.append(_Abcd(before=actual, after=before))
delta = cliffs(lst1=Bugs(predTest), lst2=after).delta()
self.out.append(delta)
if self.extent == 0:
append = 'Base'
else:
if self.Prune:
append = str(
self.extent) + '_iP(' + str(
int(self.infoPrune * 100)) + r'%)' if not self.fSelect else str(
self.extent) + '_w_iP(' + str(
int(self.infoPrune * 100)) + r'%)'
else:
append = str(
self.extent) if not self.fSelect else str(
self.extent) + '_w'
self.out.insert(0, self.dataName + '_' + append)
self.out_pred.insert(0, self.dataName)
print(self.out)
def main():
dir = '../Data'
from os import walk
dataName = [Name for _, Name, __ in walk(dir)][0]
numData = len(dataName) # Number of data
Prd = [CART] # , rforest] # , adaboost, logit, knn]
_smoteit = [True] # , False]
_tuneit = [False]
cd = {}
abcd = []
res = {}
for n in xrange(numData):
out11 = []
outA1 = []
out1 = []
outFar = []
outNear = []
outa = []
one, two = explore(dir)
data = [one[i] + two[i] for i in xrange(len(one))]
print('##', dataName[n])
for p in Prd:
train = [dat[0] for dat in withinClass(data[n])]
test = [dat[1] for dat in withinClass(data[n])]
reps = 10
abcd = [[], []]
for t in _tuneit:
tunedParams = None if not t else params
print('### Tuning') if t else print('### No Tuning')
for _smote in _smoteit:
# for _n in xrange(0):
_n = -1
# Training data
for _ in xrange(reps):
train_DF = createTbl(train[_n], isBin=True)
# set_trace()
# Testing data
test_df = createTbl(test[_n], isBin=True)
predRows = []
# Tune?
actual = Bugs(test_df)
before = p(train_DF, test_df,
tunings=tunedParams,
smoteit=True)
tunedParams = None if not t else tuner(p, train[_n])
for predicted, row in zip(before, test_df._rows):
tmp = row.cells
tmp[-2] = predicted
if predicted > 0:
predRows.append(tmp)
predTest = clone(test_df, rows=predRows)
# Find and apply contrast sets
# newTab = treatments(train = train[_n],
# test = test[_n],
# verbose = False,
# smoteit = False).main()
newTab_near = treatments2(train=train[_n], far=False, test=test[_n] # ).main()
, test_df=predTest).main() \
if predRows \
else treatments2(train=train[_n], test=test[_n]).main()
newTab_far = treatments2(train=train[_n], test=test[_n] # ).main()
, test_df=predTest).main() \
if predRows \
else treatments2(train=train[_n], test=test[_n]).main()
after_far = p(train_DF, newTab_far,
tunings=tunedParams,
smoteit=True)
after_near = p(train_DF, newTab_near,
tunings=tunedParams,
smoteit=True)
# print(showoff(dataName[n], before, after))
outa.append(_Abcd(before=actual, after=before))
# set_trace()
cliffsFar = cliffsdelta(Bugs(predTest), after_far)
cliffsNear = cliffsdelta(Bugs(predTest), after_near)
# print(cliffsDelta(Bugs(predTest), after))
# print('Gain = %1.2f' % float(\
# (sum(Bugs(predTest)) - sum(after)) / sum(Bugs(predTest)) * 100), r'%')
outFar.append(cliffsFar)
outNear.append(cliffsNear)
# out1.append(float((sum(before) - sum(after)) / sum(before) * 100))
# out1 = [o for o in out1 if np.isfinite(o)]
outNear.insert(0, dataName[n] + '_Far')
outFar.insert(0, dataName[n] + '_Near')
outa.insert(0, dataName[n])
out11.extend([outNear, outFar])
outA1.append(outa)
try:
print('```')
rdivDemo(out11, isLatex=False)
# rdivDemo(outA1, isLatex = False)
print('```')
except IndexError:
pass
def depen(self, rows):
mod = rforest(self.train, self.test
, tunings = rows # n_est, max_feat, mss, msl
, smoteit = True)
g = _Abcd(before = Bugs(self.test), after = mod, show = False)[-1]
return g
def main():
dir = '../Data'
from os import walk
dataName = [Name for _, Name, __ in walk(dir)][0]
numData = len(dataName) # Number of data
Prd = [CART] # , rforest] # , adaboost, logit, knn]
_smoteit = [False] # , False]
_tuneit = [False] # , False]
cd = []
abcd = []
res = {}
for n in xrange(numData):
one, two = explore(dir)
data = [one[i] + two[i] for i in xrange(len(one))];
print('##', dataName[n])
for p in Prd:
# print(p.__doc__)
# params = tuner(p, data[0])
# print(params)
train = [dat[0] for dat in withinClass(data[n])]
test = [dat[1] for dat in withinClass(data[n])]
reps = 1
abcd = [[], []];
for t in _tuneit:
# print('### Tuning') if t else print('### No Tuning')
for _smote in _smoteit:
# print('### SMOTE-ing') if _smote else print('### No SMOTE-ing')
# print('```')
# for _n in xrange(0):
# set_trace()
_n = -1
# Training data
for _ in xrange(reps):
train_DF = createTbl(train[_n])
# set_trace()
# Testing data
test_df = createTbl(test[_n])
# Tune?
tunedParams = None
# tunedParams = None if not t else params
# Find and apply contrast sets
newTab = treatments(train = train[_n],
test = test[_n], verbose = False, smoteit = True)
# Actual bugs
actual = Bugs(test_df)
actual1 = [0 if a == 0 else 1 for a in actual]
# Use the classifier to predict the number of bugs in the raw data.
before = p(train_DF, test_df,
tunings = tunedParams,
smoteit = _smote)
before1 = [0 if b == 0 else 1 for b in before]
# Use the classifier to predict the number of bugs in the new data.
after = p(train_DF, newTab,
tunings = tunedParams,
smoteit = _smote)
after1 = [0 if a == 0 else 1 for a in after]
# % set_trace()
# write('.')
# write('Training: '); [write(l + ', ') for l in train[_n]]; print('\n')
# cd.append(showoff(dataName[n], actual1, after1))
# print(showoff(dataName[n], actual1, after1))
# write('Test: '); [write(l) for l in test[_n]],
out = _Abcd(before = actual1, after = before1)
print('Win Ratio : %0.2d' % (sum(after1) / sum(actual1)))
# %print('Prediction accuracy (g) %.2d' % out[-1])
# print (out[-1])
if _smote:
out.insert(0, p.__doc__ + '(s, Tuned) ') if t \
else out.insert(0, p.__doc__ + '(s, Naive) ')
abcd[0].append(out)
else:
out.insert(0, p.__doc__ + '(raw, Tuned)') if t \
else out.insert(0, p.__doc__ + '(raw, Naive)')
abcd[1].append(out)
print()
# cd.update({p.__doc__:sorted(cd)})
# res.update({p.__doc__:(abcd[0][0:reps],
# abcd[0][reps:] ,
# abcd[1][0:reps],
# abcd[1][reps:] ,
# )})
print('```')
# print(cd)
# printsk(res)
print('```')
def go(self):
for _ in xrange(self.reps):
predRows = []
train_DF = createTbl(self.train[self._n][-2:], isBin=True)
test_df = createTbl(self.test[self._n], isBin=True)
actual = Bugs(test_df)
before = self.pred(train_DF,
test_df,
tunings=self.tunedParams,
smoteit=True)
for predicted, row in zip(before, test_df._rows):
tmp = row.cells
tmp[-2] = predicted
if predicted > 0:
predRows.append(tmp)
predTest = clone(test_df, rows=predRows)
if predRows:
newTab = treatments2(train=self.train[self._n][-2:],
test=self.test[self._n],
test_df=predTest,
extent=self.extent,
far=False,
smote=True,
resample=False,
infoPrune=self.infoPrune,
Prune=self.Prune).main()
else:
newTab = treatments2(train=self.train[self._n][-2:],
test=self.test[self._n],
far=False,
smote=True,
resample=False,
extent=self.extent,
infoPrune=self.infoPrune,
Prune=self.Prune).main()
after = self.pred(train_DF,
newTab,
tunings=self.tunedParams,
smoteit=True)
self.out_pred.append(_Abcd(before=actual, after=before))
delta = cliffs(lst1=Bugs(predTest), lst2=after).delta()
self.out.append(delta)
if self.extent == 0:
append = 'Base'
else:
if self.Prune:
append = str(self.extent) + '_iP(' + str(
int(self.infoPrune *
100)) + r'%)' if not self.fSelect else str(
self.extent) + '_w_iP(' + str(
int(self.infoPrune * 100)) + r'%)'
else:
append = str(self.extent) if not self.fSelect else str(
self.extent) + '_w'
self.out.insert(0, self.dataName + '_' + append)
self.out_pred.insert(0, self.dataName)
print(self.out)
def depen(self, rows): mod = CART(self.train, self.test, tunings=rows, smoteit=False) g = _Abcd(before=Bugs(self.test), after=mod, show=False)[-1] return g
def depen(self, rows):
mod = CART(self.train, self.test, tunings=rows, smoteit=True)
g = _Abcd(before=Bugs(self.test), after=mod, show=False)[-1]
return g
def main():
dir = '../Data'
from os import walk
dataName = [Name for _, Name, __ in walk(dir)][0]
numData = len(dataName) # Number of data
Prd = [CART] # , rforest] # , adaboost, logit, knn]
_smoteit = [False] # , False]
_tuneit = [False] # , False]
cd = []
abcd = []
res = {}
for n in xrange(numData):
one, two = explore(dir)
data = [one[i] + two[i] for i in xrange(len(one))]
print('##', dataName[n])
for p in Prd:
# print(p.__doc__)
# params = tuner(p, data[0])
# print(params)
train = [dat[0] for dat in withinClass(data[n])]
test = [dat[1] for dat in withinClass(data[n])]
reps = 1
abcd = [[], []]
for t in _tuneit:
# print('### Tuning') if t else print('### No Tuning')
for _smote in _smoteit:
# print('### SMOTE-ing') if _smote else print('### No SMOTE-ing')
# print('```')
# for _n in xrange(0):
# set_trace()
_n = -1
# Training data
for _ in xrange(reps):
train_DF = createTbl(train[_n])
# set_trace()
# Testing data
test_df = createTbl(test[_n])
# Tune?
tunedParams = None
# tunedParams = None if not t else params
# Find and apply contrast sets
newTab = treatments(train=train[_n],
test=test[_n],
verbose=False,
smoteit=True)
# Actual bugs
actual = Bugs(test_df)
actual1 = [0 if a == 0 else 1 for a in actual]
# Use the classifier to predict the number of bugs in the raw data.
before = p(train_DF,
test_df,
tunings=tunedParams,
smoteit=_smote)
before1 = [0 if b == 0 else 1 for b in before]
# Use the classifier to predict the number of bugs in the new data.
after = p(train_DF,
newTab,
tunings=tunedParams,
smoteit=_smote)
after1 = [0 if a == 0 else 1 for a in after]
# % set_trace()
# write('.')
# write('Training: '); [write(l + ', ') for l in train[_n]]; print('\n')
# cd.append(showoff(dataName[n], actual1, after1))
# print(showoff(dataName[n], actual1, after1))
# write('Test: '); [write(l) for l in test[_n]],
out = _Abcd(before=actual1, after=before1)
print('Win Ratio : %0.2d' %
(sum(after1) / sum(actual1)))
# %print('Prediction accuracy (g) %.2d' % out[-1])
# print (out[-1])
if _smote:
out.insert(0, p.__doc__ + '(s, Tuned) ') if t \
else out.insert(0, p.__doc__ + '(s, Naive) ')
abcd[0].append(out)
else:
out.insert(0, p.__doc__ + '(raw, Tuned)') if t \
else out.insert(0, p.__doc__ + '(raw, Naive)')
abcd[1].append(out)
print()
# cd.update({p.__doc__:sorted(cd)})
# res.update({p.__doc__:(abcd[0][0:reps],
# abcd[0][reps:] ,
# abcd[1][0:reps],
# abcd[1][reps:] ,
# )})
print('```')
# print(cd)
# printsk(res)
print('```')