Skip to content

Commit

Permalink
run benchmarks
Browse files Browse the repository at this point in the history
  • Loading branch information
@yoavg
yoavg committed May 16, 2017
1 parent 17b42a0 commit 85fd0a1
Show file tree
Hide file tree
Showing 3 changed files with 77 additions and 16 deletions.
78 changes: 62 additions & 16 deletions barchybrid/src/arc_hybrid.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,12 +5,29 @@
import utils, time, random
import numpy as np
import dynet as dy
from collections import defaultdict

import random
random.seed(1)

dy.renew_cg()

class Timer:
def __init__(self):
self.t = defaultdict(float)
self.s = 0
self.w = defaultdict(int)
def start(self):
self.s = time.time()
#print "start"
def stop(self, msg):
self.t[msg] += (time.time()-self.s)
self.w[msg] += 1
#print "stop",msg,(time.time()-self.s)
def report(self):
for k,v in self.t.iteritems():
print v,k,self.w[k],v/self.w[k]

class ArcHybridLSTM:
def __init__(self, words, pos, rels, w2i, options):
self.model = Model()
Expand Down Expand Up @@ -68,9 +85,9 @@ def __init__(self, words, pos, rels, w2i, options):
VanillaLSTMBuilder(1, self.ldims, self.ldims * 0.5, self.model)]
elif self.blstmFlag:
if self.layers > 0:
self.surfaceBuilders = [VanillaLSTMBuilder(self.layers, dims, self.ldims * 0.5, self.model), LSTMBuilder(self.layers, dims, self.ldims * 0.5, self.model)]
self.surfaceBuilders = [VanillaLSTMBuilder(self.layers, dims, self.ldims * 0.5, self.model), VanillaLSTMBuilder(self.layers, dims, self.ldims * 0.5, self.model)]
else:
self.surfaceBuilders = [SimpleRNNBuilder(1, dims, self.ldims * 0.5, self.model), LSTMBuilder(1, dims, self.ldims * 0.5, self.model)]
self.surfaceBuilders = [SimpleRNNBuilder(1, dims, self.ldims * 0.5, self.model), SimpleRNNBuilder(1, dims, self.ldims * 0.5, self.model)]

self.hidden_units = options.hidden_units
self.hidden2_units = options.hidden2_units
Expand Down Expand Up @@ -116,14 +133,10 @@ def __evaluate(self, stack, buf, train):

if self.hidden2_units > 0:
routput = (self.routLayer.expr() * self.activation(self.rhid2Bias.expr() + self.rhid2Layer.expr() * self.activation(self.rhidLayer.expr() * input + self.rhidBias.expr())) + self.routBias.expr())
else:
routput = (self.routLayer.expr() * self.activation(self.rhidLayer.expr() * input + self.rhidBias.expr()) + self.routBias.expr())

if self.hidden2_units > 0:
output = (self.outLayer.expr() * self.activation(self.hid2Bias.expr() + self.hid2Layer.expr() * self.activation(self.hidLayer.expr() * input + self.hidBias.expr())) + self.outBias.expr())
else:
output = (self.outLayer.expr() * self.activation(self.hidLayer.expr() * input + self.hidBias.expr()) + self.outBias.expr())

routput = (self.routLayer.expr() * self.activation(self.rhidLayer.expr() * input + self.rhidBias.expr()) + self.routBias.expr())
output = (self.outLayer.expr() * self.activation(self.hidLayer.expr() * input + self.hidBias.expr()) + self.outBias.expr())

return routput, output

Expand Down Expand Up @@ -174,9 +187,12 @@ def exprs_to_scores(self, routput, output, stack, buf, train):
output0 = output[0]
output1 = output[1]
output2 = output[2]
ret = [ [ (rel, 0, scrs[1 + j * 2] + uscrs1, routput[1 + j * 2 ] + output1) for j, rel in enumerate(self.irels) ] if left_arc_conditions else [],
[ (rel, 1, scrs[2 + j * 2] + uscrs2, routput[2 + j * 2 ] + output2) for j, rel in enumerate(self.irels) ] if right_arc_conditions else [],
[ (None, 2, scrs[0] + uscrs0, routput[0] + output0) ] if shift_conditions else [] ]
#ret = [ [ (rel, 0, scrs[1 + j * 2] + uscrs1, routput[1 + j * 2 ] + output1) for j, rel in enumerate(self.irels) ] if left_arc_conditions else [],
# [ (rel, 1, scrs[2 + j * 2] + uscrs2, routput[2 + j * 2 ] + output2) for j, rel in enumerate(self.irels) ] if right_arc_conditions else [],
# [ (None, 2, scrs[0] + uscrs0, routput[0] + output0) ] if shift_conditions else [] ]
ret = [ [ (rel, 0, scrs[1 + j * 2] + uscrs1, (1 + j * 2,1)) for j, rel in enumerate(self.irels) ] if left_arc_conditions else [],
[ (rel, 1, scrs[2 + j * 2] + uscrs2, (2 + j * 2,2)) for j, rel in enumerate(self.irels) ] if right_arc_conditions else [],
[ (None, 2, scrs[0] + uscrs0, (0,0)) ] if shift_conditions else [] ]
else:
s1,r1 = max(zip(scrs[1::2],self.irels))
s2,r2 = max(zip(scrs[2::2],self.irels))
Expand Down Expand Up @@ -302,7 +318,7 @@ def Predict(self, conll_path, BATCH_SIZE=5):
renew_cg()
for sent in sentence_batch: yield sent

def Train(self, conll_path, BATCH_SIZE=5):
def Train(self, conll_path, BATCH_SIZE=5, MAX_SENTS=0):
mloss = 0.0
errors = 0
batch = 0
Expand All @@ -319,12 +335,15 @@ def Train(self, conll_path, BATCH_SIZE=5):

start = time.time()

T = Timer()
st = time.time()
with open(conll_path, 'r') as conllFP:
_s = time.time()
shuffledData = list(read_conll(conllFP, True))
if MAX_SENTS>0: shuffledData = shuffledData[:MAX_SENTS]
print time.time()-_s
random.shuffle(shuffledData)
# shuffledData = [x for l,x in sorted([(-len(s),s) for s in shuffledData])][10:]

errs = []
eeloss = 0.0
Expand All @@ -334,10 +353,18 @@ def Train(self, conll_path, BATCH_SIZE=5):
nexamples=0
for iSentence, sentence_batch in enumerate(stream_to_batch(shuffledData, BATCH_SIZE),1):

T.start()
sents = [self.init_sentence(s) for s in sentence_batch]
T.stop("init")
nsents += len(sents)
nwords += sum([len(s) for s in sentence_batch])

# eval fw lstms
#print "fw:",len(sents)
T.start()
sentence_batch[-1][-1].lstms[1].forward()
T.stop("fw")

hoffset = 1 if self.headFlag else 0

while sents:
Expand All @@ -348,15 +375,25 @@ def Train(self, conll_path, BATCH_SIZE=5):
continue
new_sents.append((stack, buf))

T.start()
r,o = self.__evaluate(stack, buf, True)
T.stop("_eval")
exprs.append((r,o, stack, buf))
sents = new_sents

# run forward on all the expressions
_es = [_x[0] for _x in exprs]+[_x[1] for _x in exprs]
if _es: dy.forward(_es)
#print "fw2",len(exprs)
T.start()
if exprs: exprs[-1][1].forward()
#_es = [_x[0] for _x in exprs]+[_x[1] for _x in exprs]
#if _es: dy.forward(_es)
T.stop("fw2")
#T.stop("fw2:%s" % len(exprs))
#print "donw"
for r,o,stack,buf in exprs:
T.start()
scores = self.exprs_to_scores(r,o, stack, buf, True)
T.stop("to_scr")
scores.append([(None, 3, ninf ,None)])

alpha = stack.roots[:-2] if len(stack) > 2 else []
Expand Down Expand Up @@ -408,7 +445,10 @@ def Train(self, conll_path, BATCH_SIZE=5):
parent.lstms[bestOp + hoffset] = child.vec

if bestValid[2] < bestWrong[2] + 1.0:
loss = bestWrong[3] - bestValid[3]
vri,voi = bestValid[3]
wri,woi = bestWrong[3]
loss = (r[wri]+o[woi]) - (r[vri]+o[voi])
#loss = bestWrong[3] - bestValid[3]
mloss += 1.0 + bestWrong[2] - bestValid[2]
eloss += 1.0 + bestWrong[2] - bestValid[2]
errs.append(loss)
Expand All @@ -428,12 +468,17 @@ def Train(self, conll_path, BATCH_SIZE=5):
if errs:
eerrs = esum(errs)
_s = time.time()
T.start()
scalar_loss = eerrs.scalar_value()
T.stop("fw3")
#print "fw:", time.time()-_s, scalar_loss
_s = time.time()
T.start()
eerrs.backward()
T.stop("bw")
#print "bw:", time.time()-_s
T.start()
self.trainer.update()
T.stop("update")
else: scalar_loss = 0
errs = []
lerrs = []
Expand All @@ -456,4 +501,5 @@ def Train(self, conll_path, BATCH_SIZE=5):
print "Loss: ", mloss/nexamples
print time.time() - st
nexamples=0
T.report()
return nsents, nwords
1 change: 1 addition & 0 deletions barchybrid/src/parser.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,7 @@
parser.add_option("--dynet-mem", type="int", dest="cnn_mem", default=512)
parser.add_option("--batch-size", type="int", dest="batch_size", default=1)
parser.add_option("--dynet-autobatch", type="int", dest="dynet_autobatch", default=100)
parser.add_option("--dynet-gpu-ids", type="int", dest="gpuids", default=2)

(options, args) = parser.parse_args()
print 'Using external embedding:', options.external_embedding
Expand Down
14 changes: 14 additions & 0 deletions barchybrid/src/run_benchmarks.sh
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
#!/bin/bash

TRAIN=../../data/en-ud-train.conllu
GPUID=2

mkdir timings

#python parser.py --train $TRAIN --outdir . --params tmp.pkl --model tmp.model --epochs 1 --dynet-mem 4096 --dynet-seed 1 --userlmost --userl --bibi-lstm --batch-size 1 --dynet-autobatch 0 > timings/default.CPU.ab0
#python parser.py --train $TRAIN --outdir . --params tmp.pkl --model tmp.model --epochs 1 --dynet-mem 4096 --dynet-seed 1 --userlmost --userl --bibi-lstm --batch-size 1 --dynet-autobatch 0 --dynet-gpu-ids $GPUID > timings/default.GPU.ab0

for BATCH_SIZE in 1 2 4 8 16 32 64; do
python parser.py --train $TRAIN --outdir . --params tmp.pkl --model tmp.model --epochs 1 --dynet-mem 4096 --dynet-seed 1 --userlmost --userl --bibi-lstm --batch-size $BATCH_SIZE --dynet-autobatch 1 > timings/default.CPU.ab-$BATCH_SIZE
python parser.py --train $TRAIN --outdir . --params tmp.pkl --model tmp.model --epochs 1 --dynet-mem 4096 --dynet-seed 1 --userlmost --userl --bibi-lstm --batch-size $BATCH_SIZE --dynet-autobatch 1 --dynet-gpu-ids $GPUID > timings/default.GPU.ab-$BATCH_SIZE
done

0 comments on commit 85fd0a1

Please sign in to comment.