From 9d39440aab1bae35b45916e140a0139f715c393b Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Tue, 20 Jul 2021 16:06:11 +0200
Subject: [PATCH] tout
---
Config.py | 2 ++
Networks.py | 33 ++++++++++++++++++++++-----------
Rl.py | 21 +++++++++++++++++++++
Train.py | 2 +-
Transition.py | 18 +++++++++++-------
main.py | 39 ++++++++++++++++++++++++++++++++-------
6 files changed, 89 insertions(+), 26 deletions(-)
diff --git a/Config.py b/Config.py
index a675e4d..0aa7595 100644
--- a/Config.py
+++ b/Config.py
@@ -14,6 +14,7 @@ class Config :
self.wordIndex = 0
self.maxWordIndex = 0 #To keep a track of the max value, in case of backtrack
self.state = 0 #State of the analysis (e.g. 0=tagger, 1=parser)
+ self.nbUndone = 0 #Number of actions that has been undone and not replaced
self.stack = []
self.comments = []
self.history = []
@@ -91,6 +92,7 @@ class Config :
right = 5
print("state :", self.state, file=output)
print("stack :",[self.getAsFeature(ind, "ID") for ind in self.stack], file=output)
+ print("nbUndone :", self.nbUndone, file=output)
print("history :",[str(trans) for trans in self.history], file=output)
print("historyPop :",[(str(c[0]),"dat:"+str(c[1]),"mvt:"+str(c[2]),"reward:"+str(c[3]),"state:"+str(c[4])) for c in self.historyPop], file=output)
toPrint = []
diff --git a/Networks.py b/Networks.py
index 1c0caf4..a34561f 100644
--- a/Networks.py
+++ b/Networks.py
@@ -24,7 +24,7 @@ def createNetwork(name, dicts, outputSizes, incremental) :
elif name == "lstm" :
return LSTMNet(dicts, outputSizes, incremental)
elif name == "separated" :
- return SeparatedNet(dicts, outputSizes, incremental)
+ return SeparatedNet(dicts, outputSizes, incremental, featureFunctionAll, historyNb, historyPopNb, suffixSize, prefixSize, columns, hiddenSize)
elif name == "tagger" :
return BaseNet(dicts, outputSizes, incremental, featureFunctionNostack, historyNb, suffixSize, prefixSize, columns, hiddenSize)
@@ -188,28 +188,29 @@ class SemiNet(nn.Module):
################################################################################
class SeparatedNet(nn.Module):
- def __init__(self, dicts, outputSizes, incremental) :
+ def __init__(self, dicts, outputSizes, incremental, featureFunction, historyNb, historyPopNb, suffixSize, prefixSize, columns, hiddenSize) :
super().__init__()
self.dummyParam = nn.Parameter(torch.empty(0), requires_grad=False)
self.incremental = incremental
self.state = 0
- self.featureFunction = "b.-2 b.-1 b.0 b.1 b.2 s.0 s.1 s.2 s.0.0 s.0.-1 s.0.1 s.1.0 s.1.-1 s.1.1 s.2.0 s.2.-1 s.2.1"
- self.historyNb = 5
- self.suffixSize = 4
- self.prefixSize = 4
- self.columns = ["UPOS", "FORM"]
+ self.featureFunction = featureFunction
+ self.historyNb = historyNb
+ self.historyPopNb = historyPopNb
+ self.suffixSize = suffixSize
+ self.prefixSize = prefixSize
+ self.columns = columns
self.embSize = 64
self.nbTargets = len(self.featureFunction.split())
- self.inputSize = len(self.columns)*self.nbTargets+self.historyNb+self.suffixSize+self.prefixSize
+ self.inputSize = len(self.columns)*self.nbTargets+self.historyNb+self.historyPopNb+self.suffixSize+self.prefixSize
self.outputSizes = outputSizes
for i in range(len(outputSizes)) :
for name in dicts.dicts :
self.add_module("emb_"+name+"_"+str(i), nn.Embedding(len(dicts.dicts[name]), self.embSize))
- self.add_module("fc1_"+str(i), nn.Linear(self.inputSize * self.embSize, 1600))
- self.add_module("output_"+str(i), nn.Linear(1600, outputSizes[i]))
+ self.add_module("fc1_"+str(i), nn.Linear(self.inputSize * self.embSize, hiddenSize))
+ self.add_module("output_"+str(i), nn.Linear(hiddenSize+1, outputSizes[i]))
self.dropout = nn.Dropout(0.3)
self.apply(self.initWeights)
@@ -219,6 +220,9 @@ class SeparatedNet(nn.Module):
def forward(self, x) :
embeddings = []
+ canBack = x[...,0:1]
+ x = x[...,1:]
+
for i in range(len(self.columns)) :
embeddings.append(getattr(self, "emb_"+self.columns[i]+"_"+str(self.state))(x[...,i*self.nbTargets:(i+1)*self.nbTargets]))
y = torch.cat(embeddings,-1).view(x.size(0),-1)
@@ -227,6 +231,10 @@ class SeparatedNet(nn.Module):
historyEmb = getattr(self, "emb_HISTORY_"+str(self.state))(x[...,curIndex:curIndex+self.historyNb]).view(x.size(0),-1)
y = torch.cat([y, historyEmb],-1)
curIndex = curIndex+self.historyNb
+ if self.historyPopNb > 0 :
+ historyPopEmb = getattr(self, "emb_HISTORY_"+str(self.state))(x[...,curIndex:curIndex+self.historyPopNb]).view(x.size(0),-1)
+ y = torch.cat([y, historyPopEmb],-1)
+ curIndex = curIndex+self.historyPopNb
if self.prefixSize > 0 :
prefixEmb = getattr(self, "emb_LETTER_"+str(self.state))(x[...,curIndex:curIndex+self.prefixSize]).view(x.size(0),-1)
y = torch.cat([y, prefixEmb],-1)
@@ -237,6 +245,7 @@ class SeparatedNet(nn.Module):
curIndex = curIndex+self.suffixSize
y = self.dropout(y)
y = F.relu(self.dropout(getattr(self, "fc1_"+str(self.state))(y)))
+ y = torch.cat([y,canBack], 1)
y = getattr(self, "output_"+str(self.state))(y)
return y
@@ -251,9 +260,11 @@ class SeparatedNet(nn.Module):
def extractFeatures(self, dicts, config) :
colsValues = Features.extractColsFeatures(dicts, config, self.featureFunction, self.columns, self.incremental)
historyValues = Features.extractHistoryFeatures(dicts, config, self.historyNb)
+ historyPopValues = Features.extractHistoryPopFeatures(dicts, config, self.historyPopNb)
prefixValues = Features.extractPrefixFeatures(dicts, config, self.prefixSize)
suffixValues = Features.extractSuffixFeatures(dicts, config, self.suffixSize)
- return torch.cat([colsValues, historyValues, prefixValues, suffixValues])
+ backAction = torch.ones(1, dtype=torch.int) if Transition.Transition("BACK 1").appliable(config) else torch.zeros(1, dtype=torch.int)
+ return torch.cat([backAction, colsValues, historyValues, historyPopValues, prefixValues, suffixValues])
################################################################################
diff --git a/Rl.py b/Rl.py
index 5d04b3f..cb4d996 100644
--- a/Rl.py
+++ b/Rl.py
@@ -145,6 +145,27 @@ def rewardA(appliable, config, action, missingLinks):
return reward
################################################################################
+################################################################################
+def rewardB(appliable, config, action, missingLinks):
+ if appliable:
+ if action.name != "BACK" :
+ reward = -action.getOracleScore(config, missingLinks)
+ else :
+ canceledRewards = []
+ found = 0
+ for i in range(len(config.historyPop))[::-1] :
+ if config.historyPop[i][0].name == "NOBACK" :
+ found += 1
+ if found == action.size :
+ break
+ else :
+ canceledRewards.append(config.historyPop[i][3])
+ reward = np.log(1-sum(canceledRewards)) if -sum(canceledRewards) > 0 else -1
+ else:
+ reward = -forbiddenReward
+ return (1.0 if config.nbUndone == 0 else 2.0)*reward
+################################################################################
+
################################################################################
def rewardA2(appliable, config, action, missingLinks):
if appliable:
diff --git a/Train.py b/Train.py
index 140c896..bd20efe 100644
--- a/Train.py
+++ b/Train.py
@@ -259,7 +259,7 @@ def trainModelRl(debug, networkName, modelDir, filename, nbIter, batchSize, devF
reward = torch.FloatTensor([reward_]).to(getDevice())
newState = None
- toState = strategy[action.name][1] if action.name in strategy else -1
+ toState = strategy[fromState][action.name][1] if action.name in strategy[fromState] else -1
if appliable :
applyTransition(strategy, sentence, action, reward_)
newState = policy_net.extractFeatures(dicts, sentence).to(getDevice())
diff --git a/Transition.py b/Transition.py
index e4864a1..b977cc0 100644
--- a/Transition.py
+++ b/Transition.py
@@ -15,7 +15,7 @@ class Transition :
if len(splited) == 3 :
self.colName = splited[1]
self.argument = splited[2]
- if not self.name in ["SHIFT","REDUCE","LEFT","RIGHT","BACK","NOBACK","EOS","TAG"] :
+ if not self.name in ["SHIFT","REDUCE","LEFT","RIGHT","BACK","NOBACK","NOBACKAB","EOS","TAG"] :
raise(Exception("'%s' is not a valid transition type."%name))
def __str__(self) :
@@ -39,8 +39,9 @@ class Transition :
applyEOS(config)
elif self.name == "TAG" :
applyTag(config, self.colName, self.argument)
- elif self.name == "NOBACK" :
+ elif "NOBACK" in self.name :
data = None
+ config.nbUndone = max(0, config.nbUndone-1)
elif "BACK" in self.name :
config.historyHistory.add(str([t[0].name for t in config.historyPop]))
applyBack(config, strategy, self.size)
@@ -80,7 +81,9 @@ class Transition :
if self.name == "TAG" :
return isEmpty(config.getAsFeature(config.wordIndex, self.colName)) or config.getAsFeature(config.wordIndex, self.colName) == Dicts.Dicts.erased
if self.name == "NOBACK" :
- return True
+ return config.nbUndone == 0
+ if self.name == "NOBACKAB" :
+ return config.nbUndone != 0
if "BACK" in self.name :
if len([h[0].name for h in config.historyPop if "NOBACK" in h[0].name]) < self.size :
return False
@@ -100,7 +103,7 @@ class Transition :
return scoreOracleReduce(config, missingLinks)
if self.name == "TAG" :
return 0 if self.argument == config.getGold(config.wordIndex, self.colName) else 1
- if self.name == "NOBACK" :
+ if "NOBACK" in self.name :
return 0
if "BACK" in self.name :
return 1
@@ -182,6 +185,7 @@ def scoreOracleReduce(config, ml) :
################################################################################
def applyBack(config, strategy, size) :
i = 0
+ config.nbUndone += size+1
while True :
trans, data, movement, _, state = config.historyPop.pop()
config.moveWordIndex(-movement)
@@ -195,7 +199,7 @@ def applyBack(config, strategy, size) :
applyBackReduce(config, data)
elif trans.name == "TAG" :
applyBackTag(config, trans.colName)
- elif trans.name == "NOBACK" :
+ elif "NOBACK" in trans.name :
i += 1
else :
print("ERROR : trying to apply BACK to '%s'"%trans.name, file=sys.stderr)
@@ -301,8 +305,8 @@ def applyTag(config, colName, tag) :
################################################################################
def applyTransition(strat, config, transition, reward) :
- movement = strat[transition.name][0] if transition.name in strat else 0
- newState = strat[transition.name][1] if transition.name in strat else -1
+ movement = strat[config.state][transition.name][0] if transition.name in strat[config.state] else 0
+ newState = strat[config.state][transition.name][1] if transition.name in strat[config.state] else -1
transition.apply(config, strat)
moved = config.moveWordIndex(movement)
movement = movement if moved else 0
diff --git a/main.py b/main.py
index 38c2b87..56517ae 100755
--- a/main.py
+++ b/main.py
@@ -52,7 +52,7 @@ if __name__ == "__main__" :
parser.add_argument("--silent", "-s", default=False, action="store_true",
help="Don't print advancement infos.")
parser.add_argument("--transitions", default="eager",
- help="Transition set to use (eager | swift | tagparser | tagparserbt).")
+ help="Transition set to use (eager | swift | tagparser | tagparserbt | tagparserbt1 | recovery).")
parser.add_argument("--ts", default="",
help="Comma separated list of supplementary transitions. Example \"BACK 1,BACK 2\"")
parser.add_argument("--network", default="base",
@@ -89,7 +89,7 @@ if __name__ == "__main__" :
transitionSets = [[Transition(elem) for elem in (tagActions+args.ts.split(',')) if len(elem) > 0]]
args.predictedStr = "UPOS"
args.states = ["tagger"]
- strategy = {"TAG" : (1,0)}
+ strategy = [{"TAG" : (1,0)}]
args.network = "tagger"
probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))]]
elif args.transitions == "taggerbt" :
@@ -99,7 +99,7 @@ if __name__ == "__main__" :
transitionSets = [[Transition(elem) for elem in (tagActions+args.ts.split(',')) if len(elem) > 0], [Transition("NOBACK"), Transition("BACK 2")]]
args.predictedStr = "UPOS"
args.states = ["tagger", "backer"]
- strategy = {"TAG" : (1,1), "NOBACK" : (0,0)}
+ strategy = [{"TAG" : (1,1)}, {"NOBACK" : (0,0)}]
args.network = "tagger"
probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
[list(map(float, args.probaStateBack.split('-')[0].split(','))), list(map(float, args.probaStateBack.split('-')[1].split(',')))]]
@@ -107,7 +107,7 @@ if __name__ == "__main__" :
transitionSets = [[Transition(elem) for elem in (["SHIFT","REDUCE","LEFT","RIGHT"]+args.ts.split(',')) if len(elem) > 0]]
args.predictedStr = "HEAD"
args.states = ["parser"]
- strategy = {"RIGHT" : (1,0), "SHIFT" : (1,0), "LEFT" : (0,0), "REDUCE" : (0,0)}
+ strategy = [{"RIGHT" : (1,0), "SHIFT" : (1,0), "LEFT" : (0,0), "REDUCE" : (0,0)}]
probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))]]
elif args.transitions == "tagparser" :
tmpDicts = Dicts()
@@ -116,7 +116,7 @@ if __name__ == "__main__" :
transitionSets = [[Transition(elem) for elem in (tagActions+args.ts.split(',')) if len(elem) > 0], [Transition(elem) for elem in (["SHIFT","REDUCE","LEFT","RIGHT"]+args.ts.split(',')) if len(elem) > 0]]
args.predictedStr = "HEAD,UPOS"
args.states = ["tagger", "parser"]
- strategy = {"RIGHT" : (1,0), "SHIFT" : (1,0), "LEFT" : (0,1), "REDUCE" : (0,1), "TAG" : (0,1)}
+ strategy = [{"TAG" : (0,1)}, {"RIGHT" : (1,0), "SHIFT" : (1,0), "LEFT" : (0,1), "REDUCE" : (0,1)}]
probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))]]
elif args.transitions == "tagparserbt" :
@@ -126,16 +126,41 @@ if __name__ == "__main__" :
transitionSets = [[Transition(elem) for elem in tagActions if len(elem) > 0], [Transition(elem) for elem in ["SHIFT","REDUCE","LEFT","RIGHT"] if len(elem) > 0], [Transition("NOBACK"),Transition("BACK 2")]]
args.predictedStr = "HEAD,UPOS"
args.states = ["tagger", "parser", "backer"]
- strategy = {"RIGHT" : (1,2), "SHIFT" : (1,2), "LEFT" : (0,1), "REDUCE" : (0,1), "TAG" : (0,1), "NOBACK" : (0,0)}
+ strategy = [{"TAG" : (0,1)}, {"RIGHT" : (1,2), "SHIFT" : (1,2), "LEFT" : (0,1), "REDUCE" : (0,1)}, {"NOBACK" : (0,0)}]
probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
[list(map(float, args.probaStateBack.split('-')[0].split(','))), list(map(float, args.probaStateBack.split('-')[1].split(',')))]]
+ elif args.transitions == "recovery" :
+ tmpDicts = Dicts()
+ tmpDicts.readConllu(args.corpus, ["UPOS"], 0)
+ tagActions = ["TAG UPOS %s"%p for p in tmpDicts.getElementsOf("UPOS") if "__" not in p and not isEmpty(p)]
+ transitionSets = [[Transition(elem) for elem in tagActions if len(elem) > 0], [Transition(elem) for elem in ["SHIFT","REDUCE","LEFT","RIGHT"] if len(elem) > 0], [Transition("NOBACK"),Transition("NOBACKAB"),Transition("BACK 2")], [Transition(elem) for elem in tagActions if len(elem) > 0], [Transition(elem) for elem in ["SHIFT","REDUCE","LEFT","RIGHT"] if len(elem) > 0]]
+ args.predictedStr = "HEAD,UPOS"
+ args.states = ["tagger", "parser", "backer", "taggerReco", "parserReco"]
+ strategy = [{"TAG" : (0,1)}, {"RIGHT" : (1,2), "SHIFT" : (1,2), "LEFT" : (0,1), "REDUCE" : (0,1)}, {"NOBACK" : (0,0), "NOBACKAB" : (0,3)}, {"TAG" : (0,4)}, {"RIGHT" : (1,2), "SHIFT" : (1,2), "LEFT" : (0,4), "REDUCE" : (0,4)}]
+ probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
+ [list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
+ [list(map(float, args.probaStateBack.split('-')[0].split(','))), list(map(float, args.probaStateBack.split('-')[1].split(',')))],
+ [list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
+ [list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))]
+]
+ elif args.transitions == "tagparserbt1" :
+ tmpDicts = Dicts()
+ tmpDicts.readConllu(args.corpus, ["UPOS"], 0)
+ tagActions = ["TAG UPOS %s"%p for p in tmpDicts.getElementsOf("UPOS") if "__" not in p and not isEmpty(p)]
+ transitionSets = [[Transition(elem) for elem in tagActions if len(elem) > 0], [Transition(elem) for elem in ["SHIFT","REDUCE","LEFT","RIGHT"] if len(elem) > 0], [Transition("NOBACK"),Transition("BACK 1")]]
+ args.predictedStr = "HEAD,UPOS"
+ args.states = ["tagger", "parser", "backer"]
+ strategy = [{"TAG" : (0,1)}, {"RIGHT" : (1,2), "SHIFT" : (1,2), "LEFT" : (0,1), "REDUCE" : (0,1)}, {"NOBACK" : (0,0)}]
+ probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
+ [list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))],
+ [list(map(float, args.probaStateBack.split('-')[0].split(','))), list(map(float, args.probaStateBack.split('-')[1].split(',')))]]
elif args.transitions == "swift" :
transitionSets = [[Transition(elem) for elem in (["SHIFT"]+["LEFT "+str(n) for n in range(1,6)]+["RIGHT "+str(n) for n in range(1,6)]+args.ts.split(',')) if len(elem) > 0]]
args.predictedStr = "HEAD"
args.states = ["parser"]
- strategy = {"RIGHT" : (1,0), "SHIFT" : (1,0), "LEFT" : (0,0), "REDUCE" : (0,0)}
+ strategy = [{"RIGHT" : (1,0), "SHIFT" : (1,0), "LEFT" : (0,0), "REDUCE" : (0,0)}]
probas = [[list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))]]
else :
raise Exception("Unknown transition set '%s'"%args.transitions)
--
GitLab