def __init__(self, neighbors=1, *opt):
self.opt = opt
self.neighbors = neighbors
self.LEFT = 0
self.RIGHT = 1
Tree.__init__(self)
def __init__(
self,
state,
dynamics_layer_sizes,
reward_layer_sizes,
value_layer_sizes,
obs_dim,
act_dim,
memory_size,
batch_size,
target_num_leaves,
cross_val_ratio=0.1,
rand_state_0=(0, 0, 0.5, 0.05, 0.01),
):
self.dynamics = relu_MLP(dynamics_layer_sizes)
self.reward = relu_MLP(reward_layer_sizes)
self.value = relu_MLP(value_layer_sizes)
self.memory_train = ReplayBuffer(obs_dim, act_dim, memory_size,
batch_size)
self.memory_test = ReplayBuffer(obs_dim, act_dim, memory_size,
batch_size)
self.target_num_leaves = target_num_leaves
new_state_node = nodeData(
action_inbound=None,
state=state,
noise_state=rand_state_0,
value_est=self.value.predict(state),
step_reward=0,
)
self.T = Tree(new_state_node)
def to_tree(i, j):
if j - i == 1:
return Tree(sent[i], None, sent[i])
else:
k = backpt[i][j]
return Tree(label_table[i][j],
[to_tree(i, k), to_tree(k, j)], None)
def forward(self, sent_all):
"""
Parse all the sentences (algorithm is only run on sentences with length >= 3)
"""
sent_len3 = []
idx = []
trees, probs = [None] * len(sent_all), [None] * len(sent_all)
for i, sent in enumerate(sent_all):
sent_stripped = [s for s in sent if s not in PUNCT]
if len(sent_stripped) == 1:
trees[i] = Tree(sent_stripped[0], None, sent_stripped[0])
elif len(sent_stripped) == 2:
trees[i] = Tree('S', [
Tree(sent_stripped[0], None, sent_stripped[0]),
Tree(sent_stripped[1], None, sent_stripped[1])
], None)
else:
sent_len3.append(sent)
idx.append(i)
if len(sent_len3) > 0:
tree_all, prob_all = self.cky_len3(sent_len3)
for tree, prob, ind in zip(tree_all, prob_all, idx):
trees[ind] = tree
probs[ind] = prob
return trees, probs
def animal():
root = Tree("bird")
while True:
if not yes("Are you thinking of an animal? "):
break
tree = root
while tree.left is not None:
prompt = tree.cargo + "? "
if yes(prompt):
tree = tree.right
else:
tree = tree.left
guess = tree.cargo
prompt = "Is it a " + guess + "?"
if yes(prompt):
print("I rule")
continue
prompt = "What is the animal's name? "
animal = input(prompt)
prompt = "What question would distinguish a {0} from a {1}? "
question = input(prompt.format(animal, guess))
tree.cargo = question
prompt = "If the animal were {0} the answer would be? "
if yes(prompt.format(animal)):
tree.left = Tree(guess)
tree.right = Tree(animal)
else:
tree.left = Tree(animal)
tree.right = Tree(guess)
def set_trees(self, cell, biome):
tree_freq = ease(myrange_f(biome.tree_range))
if cell.color == 14:
cell.color = 3
height = myrange(self.config.trees_height_range)
girth = min(self.cell_size, self.cell_size) * 2 / 5
p = Vec2(self.cell_size / 2, self.cell_size / 2)
leaves = [[0, 0], [0, 0], [0, 0]]
cell.trees.append(Tree(p, height, girth, leaves))
else:
# Trees
for x in range(0, self.cell_size - self.config.trees_gap,
self.config.trees_gap):
for y in range(0, self.cell_size - self.config.trees_gap,
self.config.trees_gap):
if rnd(1) < tree_freq:
height = myrange(self.config.trees_height_range)
girth = myrange(self.config.trees_girth_range)
p = Vec2(x + rnd(self.config.trees_gap),
y + rnd(self.config.trees_gap))
leaves = [[0, 0], [0, 0], [0, 0]]
tree = Tree(p, height, girth, leaves)
cell.trees.append(tree)
tree.p = Vec2(
mid(tree.girth, tree.pos.x,
self.cell_size - tree.girth),
mid(tree.girth, tree.pos.y,
self.cell_size - tree.girth))
def main(ptb_file, results_dir, annotations_dir):
correct_patterns = set()
for root, dirs, files in os.walk(results_dir):
for f in files:
if 'correct_' not in f:
continue
for line in open(root + '/' + f):
correct_patterns.add(line.strip()[1:-1])
pattern_heads = dict()
for root, dirs, files in os.walk(annotations_dir):
for f in files:
for line in open(root + '/' + f):
if line.startswith('i'): continue # skip the header
index, pattern, head = line.strip().split('\t')
pattern_heads[pattern[1:-1]] = int(head) - 1 # 0-index the head
trees = []
text = ''
skipped = 0
for line in open(ptb_file):
if text and line[0] != ' ':
try:
trees.append(Tree.read(text))
except AttributeError:
#print text
skipped += 1
text = ''
text += line
if text:
trees.append(Tree.read(text))
good_trees = []
for tree in trees:
if tree_is_good(tree.root.children[0], pattern_heads):
#if tree_is_good(tree.root.children[0], correct_patterns):
good_trees.append(tree)
print 'Number of trees:', len(trees)
print 'Number of good trees:', len(good_trees)
print 'Skipped:', skipped
cats = error_category_counts.keys()
cats.sort(key=lambda x: error_category_counts[x])
for cat in cats:
print '%s: %d' % (cat, error_category_counts[cat])
print
errors = error_counts.keys()
errors.sort(key=lambda x: error_counts[x], reverse=True)
for error in errors[:30]:
print '%s: %d' % (error, error_counts[error])
out = open('good_trees.mrg', 'w')
for tree in good_trees:
out.write(tree.pretty())
out.write('\n\n')
out.close()
shuffle(good_trees)
num_examples = 100
out = open('marked_example_trees.mrg', 'w')
for tree in good_trees[:num_examples]:
mark_heads(tree.root.children[0], pattern_heads)
out.write(tree.pretty())
out.write('\n\n')
out.close()
def add_random(current_tree, depth=0):
if depth <= 0:
return
new_left = Tree(random.randint(0,100))
current_tree.left = new_left
add_random(new_left, depth-1)
new_right = Tree(random.randint(0,100))
current_tree.right = new_right
add_random(new_right, depth-1)
def make_class_tree(X, y):
tree = Tree()
tree.root = TreeNode()
C = np.unique(y)
nodes = {c: TreeNode() for c in C}
for i, c in enumerate(y):
node = nodes[c]
leaf = TreeLeaf(i)
node.add_child(leaf)
for node in nodes.values():
tree.root.add_child(node)
return tree
def make_class_tree(X, y):
tree = Tree()
tree.root = TreeNode()
C = np.unique(y)
nodes = {c: TreeNode() for c in C}
for i, c in enumerate(y):
node = nodes[c]
leaf = TreeLeaf(i)
node.add_child(leaf)
for node in nodes.values():
tree.root.add_child(node)
return tree
def testMegabatch(self):
print 'Loading test megabatch data'
img_dat = dict()
tree_dat = dict()
tree_rem = dict()
cur_queue = self.test_megabatch_queue.pop(0)
for i in cur_queue:
tree_rem[i] = self.data_dict[i]['n_desc']
vgg16 = self.img_feats['vgg16'][self.data_dict[i]['vgg16']]
vgg19 = self.img_feats['vgg19'][self.data_dict[i]['vgg19']]
imidx = self.data_dict[i]['img_feat_idx']
if self.data_type == 'both':
img_dat[i] = np.hstack((vgg16[imidx, :], vgg19[imidx, :]))
elif self.data_type == 'vgg16':
img_dat[i] = vgg16[imidx, :]
elif self.data_type == 'vgg19':
img_dat[i] = vgg19[imidx, :]
tree_dat[i] = []
for curtreeIDX in self.data_dict[i]['desc_idx']:
tree_dat[i].append(Tree(self.trees[curtreeIDX][1]))
# shuffle the trees
np.random.shuffle(tree_dat[i])
print 'Constructing schedule for this testing megabatch'
while len(tree_rem) >= self.minibatch_size:
sample = np.random.choice(tree_rem.keys(),
self.minibatch_size,
replace=False)
for k in sample:
tree_rem.pop(k, None)
sample = [[img_dat[x], tree_dat[x].pop(0)] for x in sample]
self.test_minibatch_queue.append(sample)
print 'Beginning testing megabatch'
def _generate_all(self, states, depth):
"""
Generates all trees up to a certain depth that are assigned a
given state by this BUTA.
:type states: tuple
:param states: A tuple of states
:type depth: int
:param depth: The maximum height of a generated tree
:rtype: generator
:return: This generator produces a series of tuples of trees.
The ith element of each tuple is a tree whose root node is
assigned the state states[i] by this BUTA. This generator
produces all possible tuples of this form
"""
for q in states:
check_is_nonterminal(q)
if depth <= 0:
return
if len(states) == 0:
yield ()
elif len(states) == 1:
for label, child_states in self._inverse_transition(states[0]):
if len(child_states) == 0:
yield (label, )
else:
for t in self._generate_all(child_states, depth - 1):
yield (Tree(label, t), )
else:
for t in self._generate_all(states[0:1], depth):
for s in self._generate_all(states[1:], depth):
yield t + s
def main():
annotations = Annotation.objects.select_related().filter(head_correct=True)
for annotation in annotations:
expansion = annotation.expansion
root = None
for i, line in enumerate(expansion.supa_example.split('\n')):
if 'ROOT' in line:
if root != None:
print 'Two roots found; skipping'
continue
root = i
if root == None:
# Sometimes the SUPA is empty; testing on 3/20/2013 showed that was
# the only time this happened
continue
tree = Tree.read(expansion.penn_example)
head_index = None
for i, child in enumerate(tree.root.children):
if root in child.terminal_indices():
head_index = i+1
break
if head_index:
annotation.head_index = head_index
annotation.save()
else:
print 'Head index not found...'
transaction.commit()
def check_trees_right(woods, screen):
#spawning trees in right side
if len(woods) < 10:
new_tree = Tree(screen, "right")
if pygame.time.get_ticks() % 10 == 0:
woods.add(new_tree)
def create_tree_4():
t = Tree(13)
import random
arr = list(range(1, 26))
random.shuffle(arr)
for x in arr:
insert(t, x)
return t
def iterative_minimax_strategy(game: Any) -> Any:
"""
Minimax strategy done with a tree file structure and stacks
"""
old_game = copy.deepcopy(game)
current_state = game.current_state
root = Tree(current_state)
stack = Stack()
stack.add(root)
while not stack.is_empty():
top = stack.remove()
game.current_state = top.value
if game.is_over(top.value):
game.current_state = top.value
if game.is_winner(top.value.get_current_player_name()):
top.score = 1
game.current_state = old_game.current_state
elif game.is_winner('p1') or game.is_winner('p2'):
top.score = -1
game.current_state = old_game.current_state
else:
top.score = 0
game.current_state = old_game.current_state
elif top.children == []:
stack.add(top)
for move in top.value.get_possible_moves():
new_state = top.value.make_move(game.str_to_move(str(move)))
trees = Tree(new_state)
trees.move = move
top.children.append(trees)
stack.add(trees)
else:
children_score = []
for child in top.children:
children_score.append(child.score * -1)
top.score = max(children_score)
for child in root.children:
if child.score * -1 == root.score:
return child.move
return old_game.current_state.get_possible_moves()[0]
def get_names_from_file(filename,tree_type=None):
"""
Returns a string containing names from the file, or a file-like object to POST
to gnrd
"""
names = None
# needs to be multipart/form-data
if tree_type in [TYPE_NEWICK, TYPE_NEXML]:
type = 'newick' if tree_type == TYPE_NEWICK else 'nexml'
n = Tree(filename, type)
labels = n.get_otu_labels()
labels = labels + n.get_node_labels()
# uniquify
labels = list(set(labels))
if None in labels:
del labels[labels.index(None)]
print "Extracted %d names from %s Tree" % (len(labels), type)
names ='\n'.join(labels)
else:
names = open(filename,'rb') # open in binary in case of PDF or Office document
return names
def simplify_trees(suite_file, outfile):
new_trees = []
tree = ''
for line in open(suite_file):
if line == '\n':
new_trees.append(convert_tree(Tree.read(tree)))
tree = ''
tree += line
out = open(outfile, 'w')
for tree in new_trees:
out.write(tree.pretty())
out.write('\n\n')
def animal():
# Start with a singleton
root = Tree("bird")
# Loop until the user quits
while True:
print()
if not yes("Are you thinking of an animal? "): break
# Walk the tree
tree = root
while tree.left is not None:
prompt = tree.cargo + "? "
if yes(prompt):
tree = tree.right
else:
tree = tree.left
# Make a guess
guess = tree.cargo
prompt = "Is it a " + guess + "? "
if yes(prompt):
print("I rule!")
continue
# Get new information
prompt = "What is the animal\'s name? "
animal = input(prompt)
prompt = "What question would distinguish a {0} from a {1}? "
question = input(prompt.format(animal, guess))
# Add new information to the tree
tree.cargo = question
prompt = "If the animal were {0} the answer would be? "
if yes(prompt.format(animal)):
tree.left = Tree(guess)
tree.right = Tree(animal)
else:
tree.left = Tree(animal)
tree.right = Tree(guess)
def replace_names_nexml(filename,mapping):
"""
Dangerously replaces the labels in a nexml file
"""
n = Tree(filename,'nexml')
n.replace_otu_labels(mapping)
n.replace_node_labels(mapping)
pieces = filename.split('.')[:1]
prefix = pieces[0] if len(pieces) >= 1 else fname
report_filename = prefix + '_clean.xml'
n.write_nexml_tree(report_filename)
def insert(tree, x, is_nil=is_nil):
assert is_bst(tree)
if is_nil(tree):
return Tree(x)
if x < tree.data:
if is_nil(tree.left):
tree.left = Tree(x)
tree.left.parent = tree
return tree.left
else:
return insert(tree.left, x)
if tree.data < x:
if is_nil(tree.right):
tree.right = Tree(x)
tree.right.parent = tree
return tree.right
else:
return insert(tree.right, x)
assert is_bst(tree)
return tree
def megabatchAdvance(self):
if self.testing:
self.testMegabatch()
return
print 'Loading megabatch data'
self.cur_megabatch += 1
img_dat = dict()
tree_dat = dict()
tree_rem = dict()
cur_queue = self.megabatch_queue.pop(0)
for i in cur_queue:
tree_rem[i] = self.data_dict[i]['n_desc']
vgg16 = self.img_feats['vgg16'][self.data_dict[i]['vgg16']]
vgg19 = self.img_feats['vgg19'][self.data_dict[i]['vgg19']]
imidx = self.data_dict[i]['img_feat_idx']
if self.data_type == 'both':
img_dat[i] = np.hstack((vgg16[imidx, :], vgg19[imidx, :]))
elif self.data_type == 'vgg16':
img_dat[i] = vgg16[imidx, :]
elif self.data_type == 'vgg19':
img_dat[i] = vgg19[imidx, :]
tree_dat[i] = []
for curtreeIDX in self.data_dict[i]['desc_idx']:
tree_dat[i].append([
self.trees[curtreeIDX][0],
Tree(self.trees[curtreeIDX][1])
])
# shuffle the trees
np.random.shuffle(tree_dat[i])
print 'Constructing schedule for this megabatch'
while len(tree_rem) >= self.minibatch_size:
sample = np.random.choice(tree_rem.keys(),
self.minibatch_size,
replace=False)
for k in sample:
tree_rem[k] -= 1
if tree_rem[k] == 0:
tree_rem.pop(k, None)
sample = [[img_dat[x], tree_dat[x].pop(0)] for x in sample]
sids = [x[1][0] for x in sample]
sample = [[x[0], x[1][1]] for x in sample]
self.minibatch_queue.append(sample)
self.minibatch_idents.append(sids)
if self.batchPerEpoch == None:
self.batchPerEpoch = len(
self.minibatch_queue) * (len(self.megabatch_queue) + 1)
print 'Beginning megabatch %i (epoch %i)' % (self.cur_megabatch,
self.cur_epoch)
def main(category, suites_dir):
tree_file = None
for root, dirs, files in os.walk('.'):
for f in files:
if 'sierra_postop' in f:
tree_file = f
if not tree_file:
print 'Could not find sierra_postop file! Exiting...'
exit(-1)
trees = []
patterns = []
text = ''
for line in open(tree_file):
if line == 'null\n':
print 'Null found!'
trees.append(None)
continue
if line == '\n':
trees.append(Tree.read(text))
text = ''
continue
text += line
new_trees = []
for i, tree in enumerate(trees):
# Clear off some of the extra processing that the SUPA pipeline adds
if tree is None:
new_trees.append('')
continue
root = tree.root.children[1]
clear_extra_labels(root)
new_trees.append(root)
outfile = '../' + suites_dir + '/' + category + '/' + category
outfile += '_PTBtrees_intermediate.mrg'
out = open(outfile, 'w')
for tree in new_trees:
if tree:
out.write(tree.pretty())
out.write('\n\n')
else:
out.write('\n')
out.close()
def constructBatch(self, batch):
minibatch = []
if type(batch) == str:
batch = eval(batch)
for i in batch:
imx = i.split('#')[0]
dcx = int(i.split('#')[1])
vgg16 = self.img_feats['vgg16'][self.data_dict[imx]['vgg16']]
vgg19 = self.img_feats['vgg19'][self.data_dict[imx]['vgg19']]
imidx = self.data_dict[imx]['img_feat_idx']
if self.data_type == 'both':
img_dat = np.hstack((vgg16[imidx, :], vgg19[imidx, :]))
elif self.data_type == 'vgg16':
img_dat = vgg16[imidx, :]
elif self.data_type == 'vgg19':
img_dat = vgg19[imidx, :]
# get tree data
desc_idx = self.data_dict[imx]['desc_idx'][dcx]
tree_dat = Tree(self.trees[desc_idx][1])
minibatch.append([img_dat, tree_dat])
return minibatch
def parse(self, tree):
"""
Given a tree, this function computes the states assigned to each
node of the tree (i.e., the "parse" of the tree). If this BUTA
is nondeterministic, a tree may have more than one parse.
:type tree: Tree
:param tree: A tree
:rtype: generator
:return: The possible parses of tree according to this BUTA.
Each parse is represented as a tree in which each node is
labelled with its state according to this BUTA
"""
if type(tree) is not Tree:
for q in self._transition(tree):
yield q
else:
symbol = get_root_label(tree)
parsed_children = [set(self.parse(t)) for t in tree]
for pc in product(*parsed_children):
child_states = tuple(get_root_label(t) for t in pc)
for q in self._transition(symbol, *child_states):
yield Tree(q, pc)
from trees import Tree, Node
def isBST(root, left=None, right=None):
if root is None:
return True
if left is not None and left.data >= root.data:
return False
if right is not None and right.data < root.data:
return False
return (isBST(root.left, left, root) and isBST(root.right, root, right))
t = Tree(5)
t.insertLeft(2)
t.insertRight(4)
root = t.getRoot()
t.inOrder(root)
print(isBST(root))
print ids
if os.path.exists('./tol.nwk'):
with open('tol.nwk') as fp:
tol = json.load(fp)['newick']
else:
tol = api.treemachine.induced_subtree(ott_ids=ids)['newick']
tree = Phylo.read(StringIO(tol), 'newick')
root_node = tree.root
def get_idx(name):
id = int(name.split('_')[-1][3:])
return ids.index(id)
def build_tree(node):
if node.is_terminal():
return TreeLeaf(get_idx(node.name))
else:
children = map(build_tree, node.clades)
node = TreeNode()
for child in children:
node.add_child(child)
return node
final_tree = Tree(root=build_tree(root_node))
with open('zoo.tree', 'wb') as fp:
pickle.dump(final_tree, fp)
with open('zoo.constraints', 'wb') as fp:
pickle.dump(list(final_tree.generate_constraints()), fp)
class Agent:
def __init__(
self,
state,
dynamics_layer_sizes,
reward_layer_sizes,
value_layer_sizes,
obs_dim,
act_dim,
memory_size,
batch_size,
target_num_leaves,
cross_val_ratio=0.1,
rand_state_0=(0, 0, 0.5, 0.05, 0.01),
):
self.dynamics = relu_MLP(dynamics_layer_sizes)
self.reward = relu_MLP(reward_layer_sizes)
self.value = relu_MLP(value_layer_sizes)
self.memory_train = ReplayBuffer(obs_dim, act_dim, memory_size,
batch_size)
self.memory_test = ReplayBuffer(obs_dim, act_dim, memory_size,
batch_size)
self.target_num_leaves = target_num_leaves
new_state_node = nodeData(
action_inbound=None,
state=state,
noise_state=rand_state_0,
value_est=self.value.predict(state),
step_reward=0,
)
self.T = Tree(new_state_node)
def act(self, state):
while self.T.num_leaves() < self.target_num_leaves:
state_node = self.T.softmax_sample_leaves()
state = state_node.state
noise_state = state_node.noise_state
noise_state, action = dampedSpringNoiseStep(state.noise_state)
state_est = state + self.dynamics.predict(np.vstack(state, action))
reward_est = self.reward.predict(np.vstack(state_est, action))
value_est = self.value.predict(state_est)
new_state_node = nodeData(
action_inbound=action,
state=state_est,
noise_state=noise_state,
value_est=value_est,
step_reward=reward_est,
)
self.T.add_state(
parent=state_node,
child=new_state_node,
)
target_node = self.T.softmax_sample_leaf_states()
next_node = self.T.step_toward_and_reroot(target_node)
action = next_node.action
def update(self, last_state, action, reward, state):
pass
current = stack.pop()
print(current.data)
current = current.right
else:
done = True
def inOrderRec2(root):
stack = deque()
tmp = root
stack.append(tmp)
while stack.__len__() > 0:
tmp = stack[-1]
if tmp.left != None:
stack.append(tmp.left)
tmp = tmp.left
data = stack.pop()
print(data.data)
if tmp.right != None:
stack.ap
t = Tree(5)
t.insertLeft(2)
t.insertRight(7)
root = t.getRoot()
inOrder(root)
inOrderRec(root)
def create_tree_3():
t = Tree(13)
for x in range(1, 10):
insert(t, x)
return t
from trees import Tree
tree = Tree()
tree.insert(9)
tree.insert(4)
tree.insert(20)
tree.insert(1)
tree.insert(6)
tree.insert(15)
tree.insert(170)
tree.insert(1)
print()
print(tree.lookup(5))
print(tree.lookup(1))
print(tree.lookup(17))
print('\nInorder ... (Shows real sequence on number line.')
tree.traverse('inorder')
print('\nPreorder')
tree.traverse('preorder')
print('\nPostorder')
tree.traverse('postorder')
print('\nLevelorder')
tree.traverse('levelorder') # breadth first
print('\n')
removed = tree.remove(4)
print(f'removed : {removed}' if removed is not None else 'Key Not Found!')
print('\nLevelorder')
return -1
height_diff = abs(rightHeight - leftHeight)
if height_diff > 1:
return -1
else:
return max(leftHeight, rightHeight) + 1
def isBalancedAlt(tree_node):
if heightAlt(tree_node) == -1:
return False
return True
if t1:
x = Tree()
x.add(5)
x.add(3)
x.add(2)
x.add(4)
print isBalancedAlt(x.root)
y = Tree()
y.add(4)
y.add(3)
y.add(5)
y.add(6)
print isBalancedAlt(y.root)
def searchDFS(graph, start):
visited = []
with open('tol.nwk') as fp:
tol = json.load(fp)['newick']
else:
tol = api.treemachine.induced_subtree(ott_ids=ids)['newick']
tree = Phylo.read(StringIO(tol), 'newick')
root_node = tree.root
def get_idx(name):
id = int(name.split('_')[-1][3:])
return ids.index(id)
def build_tree(node):
if node.is_terminal():
return TreeLeaf(get_idx(node.name))
else:
children = map(build_tree, node.clades)
node = TreeNode()
for child in children:
node.add_child(child)
return node
final_tree = Tree(root=build_tree(root_node))
with open('zoo.tree', 'wb') as fp:
pickle.dump(final_tree, fp)
with open('zoo.constraints', 'wb') as fp:
pickle.dump(list(final_tree.generate_constraints()), fp)
def main(annotation_file, category):
outfile = 'results/results.tsv'
tree_file = None
for root, dirs, files in os.walk('.'):
for f in files:
if 'sierra_postop' in f:
tree_file = f
if not tree_file:
print 'Could not find sierra_postop file! Exiting...'
exit(-1)
annotations = {}
annotation_patterns = {}
for line in open(annotation_file):
if line.startswith('index'): continue
index, pattern, head_index = line.strip().split('\t')
annotations[int(index)] = int(head_index) - 1
annotation_patterns[pattern] = int(head_index) - 1
trees = []
patterns = []
text = ''
for line in open(tree_file):
if line == 'null\n':
trees.append(None)
continue
if line == '\n':
trees.append(Tree.read(text))
text = ''
continue
text += line
count_file = '../test_suites_v2/%s/%s_tagAsParent_rules_grouped.txt' % (
category, category)
counts = []
i = 0
for line in open(count_file):
count, pattern, _ = line.split('\t')
counts.append(int(count))
# TODO: this could be better - like check the annotation file to be
# sure that the patterns match
patterns.append(pattern)
i += 1
if len(counts) != len(trees):
print 'Error! Incorrect alignment between trees and counts:'
print len(counts), len(trees)
exit(-1)
# 'count' is token count, 'num' is type count
total_count = 0
count_annotated = 0
count_correct = 0
num_patterns = 0
num_annotated = 0
num_correct = 0
errors = []
correct = []
for i, tree in enumerate(trees):
pattern = patterns[i]
num_patterns += 1
total_count += counts[i]
index = i + 1
if index not in annotations: continue
num_annotated += 1
count_annotated += counts[i]
# Clear off some of the extra processing that the SUPA pipeline adds
if tree is None:
continue
root = tree.root.children[1]
head = root.label.split('__', 1)[1]
head_index = -1
# The labels are on trees that haven't had WH-movement undone - we have
# to correct the annotations for that. This isn't perfect, but it will
# do for now.
annotated_children = len(pattern.split()) - 1
actual_children = len(root.children)
is_conjpp = 'CONJPP' in [x.label.split('__')[0] for x in root.children]
if (not is_conjpp
and actual_children == annotated_children - 1
and annotations[index] != 0):
annotations[index] = annotations[index] - 1
# Now to actually check to see what was labeled as the head
for j, child in enumerate(root.children):
child_head = child.label.split('__', 1)[1]
if child_head == head:
head_index = j + 1
if j == annotations[index]:
correct.append(patterns[i])
num_correct += 1
count_correct += counts[i]
break
else:
errors.append((patterns[i], head_index, annotations[index]+1))
percent_tested = num_annotated / num_patterns
percent_correct = num_correct / num_annotated
count_percent_annotated = count_annotated / total_count
count_percent_correct = count_correct / count_annotated
out = open(outfile, 'a')
out.write('%s\t%d\t%d\t%.3f\t%d\t%.3f\t%d\t%d\t%.3f\t%d\t%.3f\n' % (
category, num_patterns, num_annotated, percent_tested, num_correct,
percent_correct, total_count, count_annotated,
count_percent_annotated, count_correct, count_percent_correct))
error_file = open('results/errors_%s.tsv' % category, 'w')
error_file.write('pattern\tpredicted\tactual\n')
for error in errors:
error_file.write('%s\t%d\t%d\n' % error);
correct_file = open('results/correct_%s.tsv' % category, 'w')
for pattern in correct:
correct_file.write('%s\n' % pattern);
rightHeight = heightAlt(tree_root.r)
if rightHeight == -1:
return -1
height_diff = abs(rightHeight - leftHeight)
if height_diff > 1:
return -1
else:
return max(leftHeight, rightHeight) + 1
def isBalancedAlt(tree_node):
if heightAlt(tree_node) == -1:
return False
return True
if t1:
x = Tree()
x.add(5)
x.add(3)
x.add(2)
x.add(4)
print isBalancedAlt(x.root)
y = Tree()
y.add(4)
y.add(3)
y.add(5)
y.add(6)
print isBalancedAlt(y.root)
def searchDFS(graph, start):
visited = []
stack = Stack()
from trees.util import plot_tree
from trees import Tree, TreeNode, TreeLeaf
import matplotlib.pyplot as plt
if __name__ == "__main__":
leaf1 = TreeLeaf(1)
leaf2 = TreeLeaf(2)
leaf3 = TreeLeaf(3)
node1 = TreeNode()
node1.add_child(leaf1)
node1.add_child(leaf2)
node2 = TreeNode()
node2.add_child(node1)
node2.add_child(leaf3)
tree = Tree(node2)
plot_tree(tree)
plt.show()
p = leaf1.detach()
plot_tree(tree)
plt.show()
leaf3.attach(p)
plot_tree(tree)
plt.show()
