def PredictAll(self, env, sess, langIds, visited, candidates):
numActions, parentLang, mask = candidates.GetFeatures()
assert(numActions > 0)
numActionsNP = np.empty([1,1], dtype=np.int32)
numActionsNP[0,0] = numActions
#print("parentLang", numActions, parentLang.shape)
#print("mask", mask.shape, mask)
#print("linkLang", linkLang.shape, linkLang)
langsVisited = GetLangsVisited(visited, langIds, env)
#print("langsVisited", langsVisited)
(qValues, ) = sess.run([self.qValues, ],
feed_dict={self.parentLang: parentLang,
self.numActions: numActionsNP,
self.mask: mask,
self.langIds: langIds,
self.langsVisited: langsVisited})
#qValues = qValues[0]
#print("hidden3", hidden3.shape, hidden3)
#print("qValues", qValues.shape, qValues)
#print("linkSpecific", linkSpecific.shape)
#print("numSiblings", numSiblings.shape)
#print("numVisitedSiblings", numVisitedSiblings.shape)
#print("numMatchedSiblings", numMatchedSiblings.shape)
qValues = np.reshape(qValues, [1, qValues.shape[0] ])
#print(" qValues", qValues)
#print()
action = np.argmax(qValues[0, :numActions])
maxQ = qValues[0, action]
#print("newAction", action, maxQ)
return qValues, maxQ, action
def __init__(self, env, action, link, langIds, targetQ, visited,
candidates, nextVisited, nextCandidates):
self.action = action
self.link = link
self.langIds = langIds
self.targetQ = np.array(targetQ, copy=True)
if visited is not None:
self.visited = visited
self.langsVisited = GetLangsVisited(visited, langIds, env)
if candidates is not None:
self.candidates = candidates
numActions, parentLang, mask, numSiblings, numVisitedSiblings, numMatchedSiblings, parentMatched, linkLang = candidates.GetFeatures(
)
self.numActions = numActions
self.parentLang = np.array(parentLang, copy=True)
self.mask = np.array(mask, copy=True)
self.numSiblings = np.array(numSiblings, copy=True)
self.numVisitedSiblings = np.array(numVisitedSiblings, copy=True)
self.numMatchedSiblings = np.array(numMatchedSiblings, copy=True)
self.parentMatched = np.array(parentMatched, copy=True)
self.linkLang = np.array(linkLang, copy=True)
self.nextVisited = nextVisited
self.nextCandidates = nextCandidates
def PredictAll(self, env, sess, langIds, visited, candidates):
numActions, numCandidates, linkSpecific = candidates.GetMask()
#print("numActions", numActions)
#print("numCandidates", numCandidates.shape, numCandidates)
#print("linkSpecific", linkSpecific.shape, linkSpecific)
assert (numActions > 0)
numActionsArr = np.empty([1, 1])
numActionsArr[0, 0] = numActions
langsVisited = GetLangsVisited(visited, langIds, env)
#print("langsVisited", langsVisited)
(probs, logit, smNumer, smNumerSum, maxLogit, maskNum,
maskBigNeg) = sess.run(
[
self.probs, self.logit, self.smNumer, self.smNumerSum,
self.maxLogit, self.maskNum, self.maskBigNeg
],
feed_dict={
self.numCandidates: numCandidates,
self.linkSpecificInput: linkSpecific,
self.langsVisited: langsVisited,
self.numActions: numActionsArr
})
probs = np.reshape(probs, [probs.shape[1]])
try:
action = np.random.choice(self.params.NUM_ACTIONS, p=probs)
except:
print("langsVisited", probs, logit, smNumer, smNumerSum,
langsVisited)
print("probs", probs)
print("logit", logit)
print("maxLogit", maxLogit)
print("smNumer", smNumer)
print("smNumerSum", smNumerSum)
print("langsVisited", langsVisited)
print("numCandidates", numCandidates)
print("maskBigNeg", maskBigNeg)
bugger_something_went_wrong
# print("langsVisited", probs, logit, smNumer, smNumerSum, langsVisited)
# print("probs", probs)
# print("logit", logit)
# print("maxLogit", maxLogit)
# print("smNumer", smNumer)
# print("smNumerSum", smNumerSum)
# print("langsVisited", langsVisited)
# print("mask", mask)
# print("maskBigNeg", maskBigNeg)
# print()
#print("action", action, probs, logit, mask, langsVisited, numActions)
if np.random.rand(1) < .005:
print("action", action, probs, logit, numCandidates,
linkSpecific.tolist(), langsVisited, numActions)
#print()
return action
def PredictAll(self, env, sess, langIds, visited, candidates):
numActions, numCandidates, parentLang = candidates.GetMask()
#print("numActions", numActions)
#print("numCandidates", numCandidates.shape, numCandidates)
#print("parentLang", parentLang.shape, parentLang)
assert (numActions > 0)
langsVisited = GetLangsVisited(visited, langIds, env)
#print("langsVisited", langsVisited)
(probs, logit, smNumer, smNumerSum, maxLogit, maskBigNeg) = sess.run(
[
self.probs, self.logit, self.smNumer, self.smNumerSum,
self.maxLogit, self.maskBigNeg
],
feed_dict={
self.numCandidates: numCandidates,
self.parentLang: parentLang,
self.langsVisited: langsVisited
})
probs = np.reshape(probs, [probs.shape[1]])
try:
action = np.random.choice(self.params.NUM_ACTIONS, p=probs)
except:
print("langsVisited", probs, logit, smNumer, smNumerSum,
langsVisited)
print("probs", probs)
print("logit", logit)
print("maxLogit", maxLogit)
print("smNumer", smNumer)
print("smNumerSum", smNumerSum)
print("langsVisited", langsVisited)
print("numCandidates", numCandidates)
print("maskBigNeg", maskBigNeg)
bugger_something_went_wrong
# print("langsVisited", probs, logit, smNumer, smNumerSum, langsVisited)
# print("probs", probs)
# print("logit", logit)
# print("maxLogit", maxLogit)
# print("smNumer", smNumer)
# print("smNumerSum", smNumerSum)
# print("langsVisited", langsVisited)
# print("mask", mask)
# print("maskBigNeg", maskBigNeg)
# print()
#print("action", action, probs, logit, mask, langsVisited, parentLang, numActions)
if np.random.rand(1) < .005:
print("action", action, probs, logit, numCandidates, parentLang,
langsVisited, numActions)
#print()
return action
def PredictAll(self, env, sess, langIds, visited, candidates):
numActions, mask = candidates.GetFeatures()
#print("numActions", numActions)
#print("mask", mask.shape, mask)
#print("parentLang", parentLang.shape, parentLang)
assert (numActions > 0)
langsVisited = GetLangsVisited(visited, langIds, env)
#print("langsVisited", langsVisited)
(probs, logit, smNumer, smNumerSum, maxLogit,
maskBigNeg) = sess.run([
self.probs, self.logit, self.smNumer, self.smNumerSum,
self.maxLogit, self.maskBigNeg
],
feed_dict={
self.mask: mask,
self.langsVisited: langsVisited
})
probs = np.reshape(probs, [probs.shape[1]])
try:
action = np.random.choice(self.params.MAX_NODES, p=probs)
except:
print("langsVisited", probs, logit, smNumer, smNumerSum,
langsVisited)
print("probs", probs)
print("logit", logit)
print("maxLogit", maxLogit)
print("smNumer", smNumer)
print("smNumerSum", smNumerSum)
print("langsVisited", langsVisited)
print("mask", mask)
print("maskBigNeg", maskBigNeg)
dsds
# print("langsVisited", probs, logit, smNumer, smNumerSum, langsVisited)
# print("probs", probs)
# print("logit", logit)
# print("maxLogit", maxLogit)
# print("smNumer", smNumer)
# print("smNumerSum", smNumerSum)
# print("langsVisited", langsVisited)
# print("mask", mask)
# print("maskBigNeg", maskBigNeg)
# print()
#print("action", action, probs, logit, mask, langsVisited, parentLang, numActions)
if np.random.rand(1) < .005:
print("action", action, probs, logit, mask, langsVisited,
numActions)
#print()
return action
def PredictAll(self, env, sess, langIds, visited, candidates):
numActions, parentLang, mask, numSiblings, numVisitedSiblings, numMatchedSiblings, parentMatched, linkLang = candidates.GetFeatures(
)
#print("numActions", numActions)
numActionsNP = np.empty([1, 1], dtype=np.int32)
numActionsNP[0, 0] = numActions
assert (numActions > 0)
#print("parentLang", numActions, parentLang.shape)
#print("mask", mask.shape, mask)
langsVisited = GetLangsVisited(visited, langIds, env)
#print("langsVisited", langsVisited)
(probs, ) = sess.run(
[self.probs],
feed_dict={
self.parentLang: parentLang,
self.numActions: numActionsNP,
self.mask: mask,
self.numSiblings: numSiblings,
self.numVisitedSiblings: numVisitedSiblings,
self.numMatchedSiblings: numMatchedSiblings,
self.parentMatched: parentMatched,
self.linkLang: linkLang,
self.langIds: langIds,
self.langsVisited: langsVisited
})
#print("hidden3", hidden3.shape, hidden3)
#print("qValues", qValues.shape, qValues)
#print(" maxQ", maxQ.shape, maxQ)
#print(" probs", probs.shape, probs)
#print(" chosenAction", chosenAction.shape, chosenAction)
#print("linkSpecific", linkSpecific.shape)
#print("numSiblings", numSiblings.shape)
#print("numVisitedSiblings", numVisitedSiblings.shape)
#print("numMatchedSiblings", numMatchedSiblings.shape)
#print(" qValues", qValues)
#print()
probs = np.reshape(probs, [probs.shape[1]])
action = np.random.choice(probs, p=probs)
#print(" action", action)
action = np.argmax(probs == action)
#print(" action", action)
return action
def __init__(self, env, action, reward, link, langIds, visited, candidates, nextVisited, nextCandidates):
self.action = action
self.link = link
self.langIds = langIds
self.reward = reward
self.discountedReward = None
if visited is not None:
self.visited = visited
self.langsVisited = GetLangsVisited(visited, langIds, env)
if candidates is not None:
self.candidates = candidates
numActions, mask = candidates.GetFeatures()
self.numActions = numActions
self.mask = np.array(mask, copy=True)
self.nextVisited = nextVisited
self.nextCandidates = nextCandidates