def run_model():
mode = params.get('mode')
if not mode:
logger.warning(
'Не указан режим работы приложения! ключ --mode [load, prepare, train, predict, all] '
)
logger.info('Пример: python run_model.py --mode train ')
logger.info('По умолчанию режим: predict ')
mode = 'predict'
logger.info(f'Приложение запущено в режиме: { mode } ')
if mode == 'predict':
predict()
elif mode == 'prepare':
prepare()
elif mode == 'load':
load()
elif mode == 'train':
train()
elif mode == 'all':
load()
prepare()
train()
predict()
else:
logger.warning(f'Режим MODE = {mode} не распознан!')
logger.info(f'Приложение успешно завершено в режиме: {mode} ')
def index():
if request.method == 'GET':
resp = make_response(render_template('index.html'), 200)
return resp
if request.method == 'POST':
tweet = request.get_json()
feature_vector = collect_feature(tweet)
resp = make_response(jsonify(predict(feature_vector)), 200)
return resp
def postprocess(out):
# k = np.array([[1/9,1/9,1/9],[1/9,1/9,1/9],[1/9,1/9,1/9]])
for i in range(3):
# out[:,:,i] = scipy.ndimage.convolve(out[:,:,i], k, mode = 'mirror')
out[:, :, i] = scipy.ndimage.median_filter(out[:, :, i], 3, mode = 'mirror')
return out
rgb_imgs = []
for f in os.listdir(params['tst_dir']):
if params['model'] in f and f.endswith('.png'):
rgb_imgs.append(os.path.join(params['tst_dir'],f))
params['batch_size'] = len(rgb_imgs)
# Compute output of the model by specifying input rgb image
prediction = predict(model_fn, params, rgb_imgs, checkpointPath)
# create a reflector to obtain the second half of the mesh
reflector = np.array([1,1,-1])
# extract the predictions and save as off file in the same path as the input rgb images
for i in range(len(rgb_imgs)):
out = next(prediction)
if isinstance(out, dict):
output = postprocess(out['prediction'])
slice1 = np.reshape(output,[params['gi_size']*params['gi_size'],3])
slice2 = slice1 * reflector
utils.writeOff(rgb_imgs[i].replace('.png','.off'), np.concatenate((slice1,slice2),axis=0), params['gi_size'], True)
def do_GET(self):
self.send_response(200)
if self.path[:6] == '/image':
self.send_header('Content-Type', 'image/png')
self.end_headers()
qs = parse_qs(self.path.split('?')[1])
states, chars = (int(qs['states'][0]),
int(qs['chars'][0]))
image_file = '/raid/lingo/abau/random-walks/dataset-%d-%d-0/graph-image.png' % (states, chars)
with open(image_file,'rb') as f:
self.wfile.write(f.read())
elif self.path[:len('/load')] == '/load':
global decoder, extractor
self.send_header('Content-Type', 'application/json')
self.end_headers()
qs = parse_qs(self.path.split('?')[1])
states, chars, length = (int(qs['states'][0]),
int(qs['chars'][0]),
int(qs['length'][0]),
)
decoder_filename = '/raid/lingo/abau/random-walks/lstm-%d-%d-0-%d/epoch-0.pt' % (
states, chars, length
)
extractor_filename = '/raid/lingo/abau/random-walks/lstm-%d-%d-0-%d/extractor-model.pt' % (
states, chars, length
)
gaussians_filename = '/raid/lingo/abau/random-walks/lstm-%d-%d-0-%d/gaussians.json' % (
states, chars, length
)
print('Loading %s, %s' % (decoder_filename, extractor_filename))
decoder = torch.load(decoder_filename)
extractor = torch.load(extractor_filename)
''' # TODO incorporate gaussians somehow
if os.path.exists(gaussians_filename):
with open(gaussians_filename) as f:
distributions = DistributionsRecord(json.load(f), graph)
'''
distributions = None
self.wfile.write(json.dumps({'success': True}).encode('utf-8'))
return
elif self.path[:len('/interactive')] == '/interactive':
self.send_header('Content-Type', 'application/json')
self.end_headers()
qs = parse_qs(self.path.split('?')[1])
sentence = qs['sentence'][0]
sentence = torch.LongTensor(
[helpers.decoding_dict[c] for c in sentence]
).unsqueeze(0).cuda()
inp = sentence[:, :-1]
target = sentence[:, 1:]
predictions, extractions, gaussians = predict.predict(decoder, extractor, distributions, inp, target)
self.wfile.write(json.dumps({
'predictions': [pred[0].cpu().numpy().tolist() for pred in predictions],
'extractions': [ext[0].cpu().numpy().tolist() for ext in extractions]
# TODO incorporate Gaussians somehow
}).encode('utf-8'))
return
elif '?' in self.path:
self.send_header('Content-Type', 'application/json')
self.end_headers()
qs = parse_qs(self.path.split('?')[1])
states, chars, length, tlength = (int(qs['states'][0]),
int(qs['chars'][0]),
int(qs['length'][0]),
int(qs['tlength'][0]))
annotation_file = '/raid/lingo/abau/random-walks/lstm-%d-%d-0-%d/sample-annotated.json' % (states, chars, length)
samples_file = '/raid/lingo/abau/random-walks/testset-%d-%d-0/small_sample' % (states, chars)
with open(samples_file) as f:
sentences = f.read().split('\n')
with open(annotation_file) as f:
annotations = json.load(f)
sentences = [(s, i) for i, s in enumerate(sentences) if len(s) == tlength]
annotations = {j: annotations[str(i)] for j, (s, i) in enumerate(sentences)}
sentences = [c[0] for c in sentences]
self.wfile.write(json.dumps({
'sentences': sentences,
'annotations': annotations
}).encode('utf-8'))
else:
self.send_header('Content-Type', 'text/html')
self.end_headers()
with open('visualize-running.html') as f:
self.wfile.write(f.read().encode('utf-8'))
return