def balanced(env, maxDocs, params):
ret = []
visited = set()
langsVisited = {}
langsTodo = {}
startNode = env.nodes[sys.maxsize]
#print("startNode", startNode.Debug())
assert (len(startNode.links) == 1)
link = next(iter(startNode.links))
while link is not None and len(visited) < maxDocs:
node = link.childNode
if node.urlId not in visited:
#print("node", node.Debug())
visited.add(node.urlId)
if node.lang not in langsVisited:
langsVisited[node.lang] = 0
langsVisited[node.lang] += 1
if params.debug and len(visited) % 40 == 0:
print(" langsVisited", langsVisited)
for link in node.links:
#print(" ", childNode.Debug())
AddTodo(langsTodo, visited, link)
numParallelDocs = NumParallelDocs(env, visited)
ret.append(numParallelDocs)
link = PopLink(langsTodo, langsVisited, params)
return ret
def dumb(env, maxDocs, params, breadthOrDepth):
ret = []
todo = []
todo.append(env.rootNode)
visited = set()
langsVisited = {}
while len(todo) > 0 and len(visited) < maxDocs:
if breadthOrDepth == 0:
node = todo.pop(0)
else:
node = todo.pop(-1)
#print("node", node.Debug())
if node.urlId not in visited:
visited.add(node.urlId)
if node.lang not in langsVisited:
langsVisited[node.lang] = 0
langsVisited[node.lang] += 1
if params.debug and len(visited) % 40 == 0:
print(" langsVisited", langsVisited)
for link in node.links:
childNode = link.childNode
#print(" ", childNode.Debug())
todo.append(childNode)
numParallelDocs = NumParallelDocs(env, visited)
ret.append(numParallelDocs)
return ret
def randomCrawl(env, maxDocs, params):
ret = []
todo = []
todo.append(env.rootNode)
visited = set()
langsVisited = {}
while len(todo) > 0 and len(visited) < maxDocs:
idx = np.random.randint(0, len(todo))
node = todo.pop(idx)
#print("node", node.Debug())
if node.urlId not in visited:
visited.add(node.urlId)
if node.lang not in langsVisited:
langsVisited[node.lang] = 0
langsVisited[node.lang] += 1
if params.debug and len(visited) % 40 == 0:
print(" langsVisited", langsVisited)
for link in node.links:
childNode = link.childNode
#print(" ", childNode.Debug())
todo.append(childNode)
numParallelDocs = NumParallelDocs(env, visited)
ret.append(numParallelDocs)
return ret
def linkText(env, maxDocs, params):
ret = []
visited = set()
langsTodo = {}
startNode = env.nodes[sys.maxsize]
#print("startNode", startNode.Debug())
assert (len(startNode.links) == 1)
link = next(iter(startNode.links))
while link is not None and len(visited) < maxDocs:
node = link.childNode
if node.urlId not in visited:
#print("node", node.Debug())
visited.add(node.urlId)
for link in node.links:
#print(" ", childNode.Debug())
AddTodoLinkText(langsTodo, visited, link)
numParallelDocs = NumParallelDocs(env, visited)
ret.append(numParallelDocs)
link = PopLinkLinkText(langsTodo, params)
return ret
def Trajectory(env, epoch, params, sess, qns):
ret = []
visited = set()
langsVisited = np.zeros([1, 3]) # langId -> count
candidates = Candidates(params, env)
node = env.nodes[sys.maxsize]
#stopNode = env.nodes[0]
#link = Link("", 0, stopNode, stopNode)
#candidates.AddLink(link)
while True:
tmp = np.random.rand(1)
if tmp > 0.5:
qnA = qns.q[0]
qnB = qns.q[1]
else:
qnA = qns.q[1]
qnB = qns.q[0]
assert (node.urlId not in visited)
#print("node", node.Debug())
visited.add(node.urlId)
UpdateLangsVisited(langsVisited, node, params.langIds)
#print(" langsVisited", langsVisited)
candidates.AddLinks(node, visited, params)
numParallelDocs = NumParallelDocs(env, visited)
ret.append(numParallelDocs)
transition = Neural(env, params, candidates, visited, langsVisited,
sess, qnA, qnB)
if transition.nextURLId == 0:
break
else:
tmp = np.random.rand(1)
if tmp > 0.5:
corpus = qnA.corpus
else:
corpus = qnB.corpus
corpus.AddTransition(transition)
node = env.nodes[transition.nextURLId]
if len(visited) > params.maxDocs:
break
return ret
def byCrawlDate(env, maxDocs, params):
nodes = list(env.nodes.values())
print("nodes", len(nodes))
nodes.sort(key=lambda x: x.crawlDate)
ret = []
visited = set()
for node in nodes:
if len(visited) >= maxDocs:
break
if node.urlId in (0, sys.maxsize):
continue
#print(" node", node.crawlDate, type(node.crawlDate))
if node.urlId not in visited:
visited.add(node.urlId)
numParallelDocs = NumParallelDocs(env, visited)
ret.append(numParallelDocs)
return ret
def Trajectory(env, params, sess, qn, corpus, test):
ret = []
totReward = 0.0
totDiscountedReward = 0.0
discount = 1.0
startNode = env.nodes[sys.maxsize]
nextVisited = set()
nextVisited.add(startNode.urlId)
nextCandidates = Candidates(params, env)
nextCandidates.AddLinks(startNode, nextVisited, params)
nextCandidates.Group(nextVisited)
transition = Transition(env, -1, 0, None, params.langIds, None, None, nextVisited, nextCandidates)
#print("candidates", transition.nextCandidates.Debug())
if test:
mainStr = "lang:" + str(startNode.lang)
rewardStr = "rewards:"
actionStr = "actions:"
while True:
#print("candidates", transition.nextCandidates.Debug())
transition, reward = Neural(env, params, transition, sess, qn)
#print("visited", len(transition.visited))
#print("candidates", transition.nextCandidates.Debug())
#print("transition", transition.Debug())
#print()
numParallelDocs = NumParallelDocs(env, transition.visited)
ret.append(numParallelDocs)
totReward += reward
totDiscountedReward += discount * reward
discount *= params.gamma
if test:
mainStr += "->" + str(transition.link.childNode.lang)
rewardStr += "->" + str(reward)
actionStr += str(transition.action) + " "
if transition.link.childNode.alignedNode is not None:
mainStr += "*"
else:
corpus.AddTransition(transition)
if transition.nextCandidates.Count() == 0:
break
if len(transition.visited) > params.maxCrawl:
break
if test:
mainStr += " " + str(len(ret))
rewardStr += " " + str(totReward) + "/" + str(totDiscountedReward)
print(actionStr)
print(mainStr)
print(rewardStr)
return ret, totReward, totDiscountedReward
def Walk(env, params, sess, qns):
ret = []
visited = set()
langsVisited = np.zeros([1, 3]) # langId -> count
candidates = Candidates(params, env)
node = env.nodes[sys.maxsize]
#stopNode = env.nodes[0]
#link = Link("", 0, stopNode, stopNode)
#candidates.AddLink(link)
mainStr = "lang:" + str(node.lang)
rewardStr = "rewards:"
actionStr = "actions:"
i = 0
numAligned = 0
totReward = 0.0
totDiscountedReward = 0.0
discount = 1.0
while True:
qnA = qns.q[0]
assert (node.urlId not in visited)
#print("node", node.Debug())
visited.add(node.urlId)
#print("node.lang", node.lang, langsVisited.shape)
UpdateLangsVisited(langsVisited, node, params.langIds)
#print(" langsVisited", langsVisited)
candidates.AddLinks(node, visited, params)
numParallelDocs = NumParallelDocs(env, visited)
ret.append(numParallelDocs)
#print("candidates", candidates.Debug())
_, _, _, _, _, _, _, _, action, link, reward = NeuralWalk(
env, params, 0.0, candidates, visited, langsVisited, sess, qnA)
node = link.childNode
#print("action", action, qValues)
actionStr += str(action) + " "
totReward += reward
totDiscountedReward += discount * reward
mainStr += "->" + str(node.lang)
rewardStr += "->" + str(reward)
if node.alignedNode is not None:
mainStr += "*"
numAligned += 1
discount *= params.gamma
i += 1
if node.urlId == 0:
break
if len(visited) > params.maxDocs:
break
mainStr += " " + str(i)
rewardStr += " " + str(totReward) + "/" + str(totDiscountedReward)
print(actionStr)
print(mainStr)
print(rewardStr)
return ret, totReward, totDiscountedReward
