def pyrnn_to_pronn(pyrnn=None, output='en-default.pronn'):
"""
Converts a legacy python RNN to the new protobuf format. Benefits of the
new format include independence from particular python versions and no
arbitrary code execution issues inherent in pickle.
Args:
pyrnn (kraken.lib.lstm.SegRecognizer): pyrnn model
output (unicode): path of the converted HDF5 model
"""
proto = pyrnn_pb2.pyrnn()
proto.kind = 'pyrnn-bidi'
proto.ninput = pyrnn.Ni
proto.noutput = pyrnn.No
proto.codec.extend(pyrnn.codec.code2char.values())
parallel, softmax = pyrnn.lstm.nets
fwdnet, revnet = parallel.nets
revnet = revnet.net
for w in ('WGI', 'WGF', 'WGO', 'WCI', 'WIP', 'WFP', 'WOP'):
fwd_weights = getattr(fwdnet, w)
rev_weights = getattr(revnet, w)
fwd_ar = getattr(proto.fwdnet, w.lower())
rev_ar = getattr(proto.revnet, w.lower())
fwd_ar.dim.extend(fwd_weights.shape)
fwd_ar.value.extend(fwd_weights.reshape(-1).tolist())
rev_ar.dim.extend(rev_weights.shape)
rev_ar.value.extend(rev_weights.reshape(-1).tolist())
proto.softmax.w2.dim.extend(softmax.W2.shape)
proto.softmax.w2.value.extend(softmax.W2.reshape(-1).tolist())
with open(output, 'wb') as fp:
fp.write(proto.SerializeToString())
def pyrnn_to_pronn(pyrnn=None, output='en-default.pronn'):
"""
Converts a legacy python RNN to the new protobuf format. Benefits of the
new format include independence from particular python versions and no
arbitrary code execution issues inherent in pickle.
Args:
pyrnn (kraken.lib.lstm.SegRecognizer): pyrnn model
output (unicode): path of the converted HDF5 model
"""
proto = pyrnn_pb2.pyrnn()
proto.kind = 'pyrnn-bidi'
proto.ninput = pyrnn.Ni
proto.noutput = pyrnn.No
proto.codec.extend(pyrnn.codec.code2char.values())
parallel, softmax = pyrnn.lstm.nets
fwdnet, revnet = parallel.nets
revnet = revnet.net
for w in ('WGI', 'WGF', 'WGO', 'WCI', 'WIP', 'WFP', 'WOP'):
fwd_weights = getattr(fwdnet, w)
rev_weights = getattr(revnet, w)
fwd_ar = getattr(proto.fwdnet, w.lower())
rev_ar = getattr(proto.revnet, w.lower())
fwd_ar.dim.extend(fwd_weights.shape)
fwd_ar.value.extend(fwd_weights.reshape(-1).tolist())
rev_ar.dim.extend(rev_weights.shape)
rev_ar.value.extend(rev_weights.reshape(-1).tolist())
proto.softmax.w2.dim.extend(softmax.W2.shape)
proto.softmax.w2.value.extend(softmax.W2.reshape(-1).tolist())
with open(output, 'wb') as fp:
fp.write(proto.SerializeToString())
def load_pronn_model(cls, path: str):
"""
Loads an pronn model to VGSL.
"""
with open(path, 'rb') as fp:
net = pyrnn_pb2.pyrnn()
try:
net.ParseFromString(fp.read())
except Exception:
raise KrakenInvalidModelException('File does not contain valid proto msg')
if not net.IsInitialized():
raise KrakenInvalidModelException('Model incomplete')
# extract codec
codec = PytorchCodec(net.codec)
input = net.ninput
hidden = net.fwdnet.wgi.dim[0]
# extract weights
weightnames = ('wgi', 'wgf', 'wci', 'wgo', 'wip', 'wfp', 'wop')
fwd_w = []
rev_w = []
for w in weightnames:
fwd_ar = getattr(net.fwdnet, w)
rev_ar = getattr(net.revnet, w)
fwd_w.append(torch.Tensor(fwd_ar.value).view(list(fwd_ar.dim)))
rev_w.append(torch.Tensor(rev_ar.value).view(list(rev_ar.dim)))
t = torch.cat(fwd_w[:4])
weight_ih_l0 = t[:, :input+1]
weight_hh_l0 = t[:, input+1:]
t = torch.cat(rev_w[:4])
weight_ih_l0_rev = t[:, :input+1]
weight_hh_l0_rev = t[:, input+1:]
weight_lin = torch.Tensor(net.softmax.w2.value).view(list(net.softmax.w2.dim))
# build vgsl spec and set weights
nn = cls('[1,1,0,{} Lbxo{} O1ca{}]'.format(input, hidden, len(net.codec)))
nn.nn.L_0.layer.weight_ih_l0 = torch.nn.Parameter(weight_ih_l0)
nn.nn.L_0.layer.weight_hh_l0 = torch.nn.Parameter(weight_hh_l0)
nn.nn.L_0.layer.weight_ih_l0_reverse = torch.nn.Parameter(weight_ih_l0_rev)
nn.nn.L_0.layer.weight_hh_l0_reverse = torch.nn.Parameter(weight_hh_l0_rev)
nn.nn.L_0.layer.weight_ip_l0 = torch.nn.Parameter(fwd_w[4])
nn.nn.L_0.layer.weight_fp_l0 = torch.nn.Parameter(fwd_w[5])
nn.nn.L_0.layer.weight_op_l0 = torch.nn.Parameter(fwd_w[6])
nn.nn.L_0.layer.weight_ip_l0_reverse = torch.nn.Parameter(rev_w[4])
nn.nn.L_0.layer.weight_fp_l0_reverse = torch.nn.Parameter(rev_w[5])
nn.nn.L_0.layer.weight_op_l0_reverse = torch.nn.Parameter(rev_w[6])
nn.nn.O_1.lin.weight = torch.nn.Parameter(weight_lin)
nn.add_codec(codec)
return nn
def load_pronn_model(cls, path: str):
"""
Loads an pronn model to VGSL.
"""
with open(path, 'rb') as fp:
net = pyrnn_pb2.pyrnn()
try:
net.ParseFromString(fp.read())
except Exception:
raise KrakenInvalidModelException('File does not contain valid proto msg')
if not net.IsInitialized():
raise KrakenInvalidModelException('Model incomplete')
# extract codec
codec = PytorchCodec(net.codec)
input = net.ninput
hidden = net.fwdnet.wgi.dim[0]
# extract weights
weightnames = ('wgi', 'wgf', 'wci', 'wgo', 'wip', 'wfp', 'wop')
fwd_w = []
rev_w = []
for w in weightnames:
fwd_ar = getattr(net.fwdnet, w)
rev_ar = getattr(net.revnet, w)
fwd_w.append(torch.Tensor(fwd_ar.value).view(list(fwd_ar.dim)))
rev_w.append(torch.Tensor(rev_ar.value).view(list(rev_ar.dim)))
t = torch.cat(fwd_w[:4])
weight_ih_l0 = t[:, :input+1]
weight_hh_l0 = t[:, input+1:]
t = torch.cat(rev_w[:4])
weight_ih_l0_rev = t[:, :input+1]
weight_hh_l0_rev = t[:, input+1:]
weight_lin = torch.Tensor(net.softmax.w2.value).view(list(net.softmax.w2.dim))
# build vgsl spec and set weights
nn = cls('[1,1,0,{} Lbxo{} O1ca{}]'.format(input, hidden, len(net.codec)))
nn.nn.L_0.layer.weight_ih_l0 = torch.nn.Parameter(weight_ih_l0)
nn.nn.L_0.layer.weight_hh_l0 = torch.nn.Parameter(weight_hh_l0)
nn.nn.L_0.layer.weight_ih_l0_reverse = torch.nn.Parameter(weight_ih_l0_rev)
nn.nn.L_0.layer.weight_hh_l0_reverse = torch.nn.Parameter(weight_hh_l0_rev)
nn.nn.L_0.layer.weight_ip_l0 = torch.nn.Parameter(fwd_w[4])
nn.nn.L_0.layer.weight_fp_l0 = torch.nn.Parameter(fwd_w[5])
nn.nn.L_0.layer.weight_op_l0 = torch.nn.Parameter(fwd_w[6])
nn.nn.L_0.layer.weight_ip_l0_reverse = torch.nn.Parameter(rev_w[4])
nn.nn.L_0.layer.weight_fp_l0_reverse = torch.nn.Parameter(rev_w[5])
nn.nn.L_0.layer.weight_op_l0_reverse = torch.nn.Parameter(rev_w[6])
nn.nn.O_1.lin.weight = torch.nn.Parameter(weight_lin)
nn.add_codec(codec)
return nn
def load_pronn(fname):
"""
Loads a legacy pyrnn model in protobuf format and instantiates a
kraken.lib.lstm.SeqRecognizer object.
Args:
fname (str): Path to the protobuf file
Returns:
A kraken.lib.lstm.SeqRecognizer object
"""
logger.info(u'Trying to load prornn model from {}'.format(fname))
with open(fname, 'rb') as fp:
logger.debug(u'Initializing protobuf message')
proto = pyrnn_pb2.pyrnn()
try:
proto.ParseFromString(fp.read())
except:
logger.debug(u'File does not contain valid proto msg')
raise KrakenInvalidModelException(
'File does not contain valid proto msg')
if not proto.IsInitialized():
logger.debug(u'Message in file incomplete')
raise KrakenInvalidModelException('Model incomplete')
# extract codec
logger.debug(u'Extracting codec')
codec = kraken.lib.lstm.Codec().init(proto.codec)
hiddensize = proto.fwdnet.wgi.dim[0]
# next build a line estimator
logger.debug(u'Add line estimator')
lnorm = kraken.lib.lineest.CenterNormalizer(proto.ninput)
network = kraken.lib.lstm.SeqRecognizer(
lnorm.target_height,
hiddensize,
codec=codec,
normalize=kraken.lib.lstm.normalize_nfkc)
logger.debug(u'Setting weights on BIDILSTM')
parallel, softmax = network.lstm.nets
fwdnet, revnet = parallel.nets
revnet = revnet.net
for w in ('WGI', 'WGF', 'WGO', 'WCI', 'WIP', 'WFP', 'WOP'):
fwd_ar = getattr(proto.fwdnet, w.lower())
rev_ar = getattr(proto.revnet, w.lower())
setattr(fwdnet, w, numpy.array(fwd_ar.value).reshape(fwd_ar.dim))
setattr(revnet, w, numpy.array(rev_ar.value).reshape(rev_ar.dim))
softmax.W2 = numpy.array(proto.softmax.w2.value).reshape(
proto.softmax.w2.dim)
return network
def load_pronn(fname):
"""
Loads a legacy pyrnn model in protobuf format and instantiates a
kraken.lib.lstm.SeqRecognizer object.
Args:
fname (unicode): Path to the HDF5 file
Returns:
A kraken.lib.lstm.SeqRecognizer object
"""
with open(fname, 'rb') as fp:
proto = pyrnn_pb2.pyrnn()
try:
proto.ParseFromString(fp.read())
except:
raise KrakenInvalidModelException(
'File does not contain valid proto msg')
if not proto.IsInitialized():
raise KrakenInvalidModelException('Model incomplete')
# extract codec
codec = kraken.lib.lstm.Codec().init(proto.codec)
hiddensize = proto.fwdnet.wgi.dim[0]
# next build a line estimator
lnorm = kraken.lib.lineest.CenterNormalizer(proto.ninput)
network = kraken.lib.lstm.SeqRecognizer(
lnorm.target_height,
hiddensize,
codec=codec,
normalize=kraken.lib.lstm.normalize_nfkc)
parallel, softmax = network.lstm.nets
fwdnet, revnet = parallel.nets
revnet = revnet.net
for w in ('WGI', 'WGF', 'WGO', 'WCI', 'WIP', 'WFP', 'WOP'):
fwd_ar = getattr(proto.fwdnet, w.lower())
rev_ar = getattr(proto.revnet, w.lower())
setattr(fwdnet, w, numpy.array(fwd_ar.value).reshape(fwd_ar.dim))
setattr(revnet, w, numpy.array(rev_ar.value).reshape(rev_ar.dim))
softmax.W2 = numpy.array(proto.softmax.w2.value).reshape(
proto.softmax.w2.dim)
return network
def load_pronn(fname):
"""
Loads a legacy pyrnn model in protobuf format and instantiates a
kraken.lib.lstm.SeqRecognizer object.
Args:
fname (unicode): Path to the HDF5 file
Returns:
A kraken.lib.lstm.SeqRecognizer object
"""
with open(fname, 'rb') as fp:
proto = pyrnn_pb2.pyrnn()
try:
proto.ParseFromString(fp.read())
except:
raise KrakenInvalidModelException('File does not contain valid proto msg')
if not proto.IsInitialized():
raise KrakenInvalidModelException('Model incomplete')
# extract codec
codec = kraken.lib.lstm.Codec().init(proto.codec)
hiddensize = proto.fwdnet.wgi.dim[0]
# next build a line estimator
lnorm = kraken.lib.lineest.CenterNormalizer(proto.ninput)
network = kraken.lib.lstm.SeqRecognizer(lnorm.target_height,
hiddensize,
codec=codec,
normalize=kraken.lib.lstm.normalize_nfkc)
parallel, softmax = network.lstm.nets
fwdnet, revnet = parallel.nets
revnet = revnet.net
for w in ('WGI', 'WGF', 'WGO', 'WCI', 'WIP', 'WFP', 'WOP'):
fwd_ar = getattr(proto.fwdnet, w.lower())
rev_ar = getattr(proto.revnet, w.lower())
setattr(fwdnet, w, numpy.array(fwd_ar.value).reshape(fwd_ar.dim))
setattr(revnet, w, numpy.array(rev_ar.value).reshape(rev_ar.dim))
softmax.W2 = numpy.array(proto.softmax.w2.value).reshape(proto.softmax.w2.dim)
return network
