def task(self):
listInst = dtree.load_csv_dataset(datadir("data.csv"))
dt = dtree.build_tree(listInst[:-10])
dtree.prune_tree(dt, listInst[-10:])
return serialize_tree(dt)
def test_prune_tree(self):
"""
Test bottom-up pruning with a validation set.
The test builds a random tree, then randomly chooses a node at which
to prune. To induce pruning, the test does the following:
- set the default label of the node to T
- set the default label of the nodes, and actual label of the leaves,
of all descendants to F
- generate a large number of T instances that follow a path
through the node
- set the default labels of all ancestors of the node to F
- prune the tree
- repeat for the node's parent, continuing up to the root.
"""
def set_labels(dtRoot,f):
def down(dt):
if dt.is_leaf():
dt.fLabel = f
dt.fDefaultLabel = f
map(down,dt.dictChildren.values())
down(dtRoot)
def check_passes(dtRoot,dtCheck,inst):
def down(dt):
assert not dt.is_leaf()
assert len(dt.dictChildren) == cValue
dt = dt.dictChildren[inst.listAttrs[dt.ixAttr]]
if dt == dtCheck:
return
down(dtRoot)
cAttr = random.randint(2,4)
cValue = random.randint(2,4)
dtBase = build_random_tree(cAttr,cValue)
listPath = []
listAttrs = []
listDt = []
fTargetValue = True#randbool()
set_labels(dtBase, not fTargetValue)
dt = dtBase
while not dt.is_leaf():
ixValue = random.choice(dt.dictChildren.keys())
listPath.append(ixValue)
listAttrs.append(dt.ixAttr)
dt = dt.dictChildren[ixValue]
while listPath:
listPath.pop()
dt = dtRoot = dtBase.copy()
for ixValue in listPath:
dt = dt.dictChildren[ixValue]
assert dt.is_node()
dt.fDefaultLabel = fTargetValue
listInst = []
fxnEnd = lambda: randlist(0,cValue-1,cAttr - len(listPath))
for _ in xrange(random.randint(1,10)):
listValue = listPath + fxnEnd()
listInstAttr = [None for _ in xrange(cAttr)]
assert len(listValue) == cAttr
for ixValue,ixAttr in zip(listValue,listAttrs):
listInstAttr[ixAttr] = ixValue
inst = dtree.Instance(listInstAttr, fTargetValue)
check_passes(dtRoot,dt,inst)
listInst.append(inst)
dtree.prune_tree(dtRoot,listInst)
dt = dtRoot
for ix,ixValue in enumerate(listPath):
assert dt.ixAttr == listAttrs[ix]
self.assertTrue(dt.is_node(), str(dtRoot))
self.assertTrue(ixValue in dt.dictChildren)
dt = dt.dictChildren[ixValue]
self.assertTrue(dt.is_leaf(), str(dt))
def test_prune_tree(self):
"""
Test bottom-up pruning with a validation set.
The test builds a random tree, then randomly chooses a node at which
to prune. To induce pruning, the test does the following:
- set the default label of the node to T
- set the default label of the nodes, and actual label of the leaves,
of all descendants to F
- generate a large number of T instances that follow a path
through the node
- set the default labels of all ancestors of the node to F
- prune the tree
- repeat for the node's parent, continuing up to the root.
"""
def set_labels(dtRoot, f):
def down(dt):
if dt.is_leaf():
dt.fLabel = f
dt.fDefaultLabel = f
map(down, dt.dictChildren.values())
down(dtRoot)
def check_passes(dtRoot, dtCheck, inst):
def down(dt):
assert not dt.is_leaf()
assert len(dt.dictChildren) == cValue
dt = dt.dictChildren[inst.listAttrs[dt.ixAttr]]
if dt == dtCheck:
return
down(dtRoot)
cAttr = random.randint(2, 4)
cValue = random.randint(2, 4)
dtBase = build_random_tree(cAttr, cValue)
listPath = []
listAttrs = []
fTargetValue = True # randbool()
set_labels(dtBase, not fTargetValue)
dt = dtBase
while not dt.is_leaf():
ixValue = random.choice(dt.dictChildren.keys())
listPath.append(ixValue)
listAttrs.append(dt.ixAttr)
# print ixValue
dt = dt.dictChildren[ixValue]
# print "-----------------------"
while listPath:
listPath.pop()
dt = dtRoot = dtBase
for ixValue in listPath:
# print ixValue
dt = dt.dictChildren[ixValue]
assert dt.is_node()
# print "-----------------------------------"
dt.fDefaultLabel = fTargetValue
listInst = []
fxnEnd = lambda: randlist(0, cValue - 1, cAttr - len(listPath))
for _ in xrange(random.randint(1, 10)):
listValue = listPath + fxnEnd()
listInstAttr = [None for _ in xrange(cAttr)]
assert len(listValue) == cAttr
for ixValue, ixAttr in zip(listValue, listAttrs):
listInstAttr[ixAttr] = ixValue
inst = dtree.Instance(listInstAttr, fTargetValue)
check_passes(dtRoot, dt, inst)
listInst.append(inst)
dtree.prune_tree(dtRoot, listInst)
dt = dtRoot
for ix, ixValue in enumerate(listPath):
assert dt.ixAttr == listAttrs[ix]
self.assertTrue(dt.is_node(), str(dtRoot))
self.assertTrue(ixValue in dt.dictChildren)
dt = dt.dictChildren[ixValue]
self.assertTrue(dt.is_leaf(), str(dt))
def task(self):
listInst = dtree.load_csv_dataset(datadir("data.csv"))
dt = dtree.build_tree(listInst[:-10])
dtree.prune_tree(dt, listInst[-10:])
return serialize_tree(dt)
