def test_gru_model_output(self):
folder_name, dataset_name, args = generate_dataset_and_parser()
execute_train(model_class=GRU,
model_args=dict(recurrent_layers=1,
readout_layers=1,
hidden_size=5),
args=args)
assert path.exists('GRU.pkl')
assert path.exists('0.pkl') or path.exists('1.pkl')
os.remove('GRU.pkl')
remove_files(folder_name, dataset_name)
def test_mlp_model_output(self):
folder_name, dataset_name, args = generate_dataset_and_parser()
execute_train(model_class=MLPEncoder,
model_args=dict(layers=2,
hidden_size=5,
batch_norm=True),
args=args)
assert path.exists('MLPEncoder.pkl')
assert path.exists('0.pkl') or path.exists('1.pkl')
os.remove('MLPEncoder.pkl')
remove_files(folder_name, dataset_name)
def test_hyperbolic_distance_output(self):
folder_name, dataset_name, args = generate_dataset_and_parser()
args = copy.copy(args)
args.distance = 'hyperbolic'
args.scaling = True
execute_train(model_class=MLPEncoder,
model_args=dict(layers=2,
hidden_size=5,
batch_norm=True),
args=args)
assert path.exists('MLPEncoder.pkl')
assert path.exists('0.pkl') or path.exists('1.pkl')
os.remove('MLPEncoder.pkl')
remove_files(folder_name, dataset_name)
def test_transformer_model_output(self):
folder_name, dataset_name, args = generate_dataset_and_parser()
execute_train(model_class=Transformer,
model_args=dict(segment_size=2,
trans_layers=1,
readout_layers=1,
hidden_size=4,
mask='empty',
heads=2,
layer_norm=True),
args=args)
assert path.exists('Transformer.pkl')
assert path.exists('0.pkl') or path.exists('1.pkl')
os.remove('Transformer.pkl')
remove_files(folder_name, dataset_name)
def test_cnn_model_output(self):
folder_name, dataset_name, args = generate_dataset_and_parser()
execute_train(model_class=CNN,
model_args=dict(readout_layers=1,
channels=4,
layers=2,
kernel_size=3,
pooling='avg',
non_linearity=True,
batch_norm=True,
stride=1),
args=args)
assert path.exists('CNN.pkl')
assert path.exists('0.pkl') or path.exists('1.pkl')
os.remove('CNN.pkl')
remove_files(folder_name, dataset_name)
def test_sequence_output(self):
folder_name, edit_dataset_name, args = generate_dataset_and_parser()
args = copy.copy(args)
args.distance = 'cosine'
# run method storing output
execute_train(model_class=MLPEncoder,
model_args=dict(layers=2,
hidden_size=5,
batch_norm=True),
args=args)
# check output tree
assert path.exists('sequences.fasta'), "Sequences file missing"
# remove files
remove_files(folder_name, edit_dataset_name)
def test_euclidean_distance_stdout(self):
folder_name, edit_dataset_name, closest_dataset_name, args = generate_dataset_and_parser()
args = copy.copy(args)
args.distance = 'euclidean'
# run method storing output
f = io.StringIO()
with redirect_stdout(f):
execute_train(model_class=MLPEncoder,
model_args=dict(layers=2,
hidden_size=5,
batch_norm=True),
args=args)
out = f.getvalue()
# check correct output
assert 'Top1:' in out and 'Top5:' in out and 'Top10:' in out, 'Wrong output format for euclidean distance'
# remove files
remove_files(folder_name, edit_dataset_name, closest_dataset_name)
def test_cosine_distance_stdout(self):
folder_name, edit_dataset_name, hc_dataset_name, args = generate_dataset_and_parser(
)
args = copy.copy(args)
args.distance = 'cosine'
# run method storing output
f = io.StringIO()
with redirect_stdout(f):
execute_train(model_class=MLPEncoder,
model_args=dict(layers=2,
hidden_size=5,
batch_norm=True),
args=args)
out = f.getvalue()
# check correct output
assert "'single': {'DC'" in out and "'complete': {'DC'" in out and "'average': {'DC'" in out, \
'Wrong output format for cosine distance'
# remove files
remove_files(folder_name, edit_dataset_name, hc_dataset_name)
from edit_distance.models.feedforward.model import MLPEncoder
from edit_distance.train import execute_train, general_arg_parser
from util.data_handling.data_loader import BOOL_CHOICE
parser = general_arg_parser()
parser.add_argument('--layers', type=int, default=1, help='Number of fully connected layers')
parser.add_argument('--hidden_size', type=int, default=100, help='Size of hidden layers')
parser.add_argument('--batch_norm', type=str, default='False', help='Batch normalization')
args = parser.parse_args()
assert args.batch_norm in BOOL_CHOICE, "Boolean values have to be either 'True' or 'False' "
execute_train(model_class=MLPEncoder,
model_args=dict(layers=args.layers,
hidden_size=args.hidden_size,
batch_norm=True if args.batch_norm == 'True' else False),
args=args)
from edit_distance.models.transformer.model import Transformer
from edit_distance.train import execute_train, general_arg_parser
from util.data_handling.data_loader import BOOL_CHOICE
parser = general_arg_parser()
parser.add_argument('--trans_layers', type=int, default=1, help='Number of attention layers')
parser.add_argument('--readout_layers', type=int, default=1, help='Number of final fully connected layers')
parser.add_argument('--hidden_size', type=int, default=16, help='Size of hidden tokens')
parser.add_argument('--mask', type=str, default="empty", help='Whether to apply a mask (empty or local{N})')
parser.add_argument('--heads', type=int, default=1, help='Number of attention heads at each layer')
parser.add_argument('--layer_norm', type=str, default='True', help='Layer normalization')
parser.add_argument('--segment_size', type=int, default=5, help='Number of elements for each string token')
args = parser.parse_args()
assert args.layer_norm in BOOL_CHOICE, "Boolean values have to be either 'True' or 'False' "
execute_train(model_class=Transformer,
model_args=dict(segment_size=args.segment_size,
trans_layers=args.trans_layers,
readout_layers=args.readout_layers,
hidden_size=args.hidden_size,
mask=args.mask,
heads=args.heads,
layer_norm=True if args.layer_norm == 'True' else False),
args=args)
default=3,
help='Kernel size in convolutions')
parser.add_argument('--pooling',
type=str,
default='avg',
help='Pooling type (avg, max or none')
parser.add_argument('--non_linearity',
type=str,
default='True',
help='Whether to apply non-linearity to convolutions')
parser.add_argument('--batch_norm',
type=str,
default='True',
help='Batch normalization')
args = parser.parse_args()
assert args.non_linearity in BOOL_CHOICE and args.batch_norm in BOOL_CHOICE, \
"Boolean values have to be either 'True' or 'False' "
execute_train(
model_class=CNN,
model_args=dict(
readout_layers=args.readout_layers,
channels=args.channels,
layers=args.layers,
kernel_size=args.kernel_size,
pooling=args.pooling,
non_linearity=True if args.non_linearity == 'True' else False,
batch_norm=True if args.batch_norm == 'True' else False,
stride=args.stride),
args=args)
from edit_distance.models.kmer.model import Kmer
from edit_distance.train import execute_train, general_arg_parser
parser = general_arg_parser()
parser.add_argument('--k', type=int, default=4, help='Size of kernel')
args = parser.parse_args()
execute_train(model_class=Kmer, model_args=dict(k=args.k), args=args)
from edit_distance.models.recurrent.model import GRU
from edit_distance.train import execute_train, general_arg_parser
parser = general_arg_parser()
parser.add_argument('--recurrent_layers',
type=int,
default=1,
help='Number of recurrent layers')
parser.add_argument('--readout_layers',
type=int,
default=2,
help='Number of readout layers')
parser.add_argument('--hidden_size',
type=int,
default=100,
help='Size of hidden dimension')
args = parser.parse_args()
execute_train(model_class=GRU,
model_args=dict(recurrent_layers=args.recurrent_layers,
readout_layers=args.readout_layers,
hidden_size=args.hidden_size),
args=args)
