def eval_bestimg(modelpath, testpath, tokenize=phonemes):
rows = [json.loads(line) for line in open(testpath)]
model = visual.load(path=modelpath)
scaler = model.scaler
batcher = model.batcher
mapper = batcher.mapper
img_fs = {}
sent_ids = {}
prov = dp.getDataProvider('coco', root='/home/gchrupala/repos/reimaginet')
for split in ['val', 'test', 'restval']:
for img in prov.iterImages(split=split):
img_fs[img['cocoid']] = scaler.transform([img['feat']])[0]
for sent in img['sentences']:
sent_ids[sent['sentid']] = sent
def response(row):
sent = sent_ids[row['meta']['id']]
inputs = list(mapper.transform([tokenize(sent)]))
pred = model.Visual.predict(batcher.batch_inp(inputs))[0]
return 1 + numpy.argmin([
cosine(pred, img_fs[cocoid])
for cocoid in row['meta']['candidates']
])
preds = numpy.array([response(row) for row in rows])
target = numpy.array([row['meta']['response'] for row in rows])
return numpy.mean(preds == target)
def evaluate(dataset='coco',
datapath='.',
model_path='model.zip',
batch_size=128,
tokenize=phonemes
):
model = visual.load(path=model_path)
task = model.Visual
scaler = model.scaler
batcher = model.batcher
mapper = batcher.mapper
prov = dp.getDataProvider(dataset, root=datapath)
sents_tok = [ tokenize(sent) for sent in prov.iterSentences(split='val') ]
predictions = visual.predict_img(model, sents_tok, batch_size=batch_size)
sents = list(prov.iterSentences(split='val'))
images = list(prov.iterImages(split='val'))
img_fs = list(scaler.transform([ image['feat'] for image in images ]))
correct_img = numpy.array([ [ sents[i]['imgid']==images[j]['imgid']
for j in range(len(images)) ]
for i in range(len(sents)) ] )
return ranking(img_fs, predictions, correct_img, ns=(1,5,10), exclude_self=False)
def eval_bestimg(modelpath, testpath, tokenize=phonemes):
rows = [ json.loads(line) for line in open(testpath)]
model = visual.load(path=modelpath)
scaler = model.scaler
batcher = model.batcher
mapper = batcher.mapper
img_fs = {}
sent_ids = {}
prov = dp.getDataProvider('coco', root='/home/gchrupala/repos/reimaginet')
for split in ['val','test','restval']:
for img in prov.iterImages(split=split):
img_fs[img['cocoid']] = scaler.transform([ img['feat'] ])[0]
for sent in img['sentences']:
sent_ids[sent['sentid']]=sent
def response(row):
sent = sent_ids[row['meta']['id']]
inputs = list(mapper.transform([tokenize(sent) ]))
pred = model.Visual.predict(batcher.batch_inp(inputs))[0]
return 1+numpy.argmin([ cosine(pred, img_fs[cocoid]) for cocoid in row['meta']['candidates']])
preds = numpy.array([ response(row) for row in rows ])
target = numpy.array([ row['meta']['response'] for row in rows])
return numpy.mean(preds==target)
def evaluate(dataset='coco',
datapath='.',
model_path='model.zip',
batch_size=128,
tokenize=phonemes):
model = visual.load(path=model_path)
task = model.Visual
scaler = model.scaler
batcher = model.batcher
mapper = batcher.mapper
prov = dp.getDataProvider(dataset, root=datapath)
sents_tok = [tokenize(sent) for sent in prov.iterSentences(split='val')]
predictions = visual.predict_img(model, sents_tok, batch_size=batch_size)
sents = list(prov.iterSentences(split='val'))
images = list(prov.iterImages(split='val'))
img_fs = list(scaler.transform([image['feat'] for image in images]))
correct_img = numpy.array(
[[sents[i]['imgid'] == images[j]['imgid'] for j in range(len(images))]
for i in range(len(sents))])
return ranking(img_fs,
predictions,
correct_img,
ns=(1, 5, 10),
exclude_self=False)