def test_save_repr():
'''
Test save repr
'''
model = initialize_model('CoCoNet', TEST_SHAPES, TEST_ARCHITECTURE)
fasta = generate_fasta_file(*TEST_CTG_LENGTHS)
coverage = generate_h5_file(*TEST_CTG_LENGTHS, filename='coverage.h5')
output = {k: Path('repr_{}.h5'.format(k))
for k in ['composition', 'coverage']}
save_repr_all(model, fasta, coverage, n_frags=5, frag_len=FL, output=output,
min_ctg_len=0, wsize=WSIZE, wstep=WSTEP)
assert all(out.is_file() for out in output.values())
handles = {k: h5py.File(v, 'r') for (k, v) in output.items()}
firsts = {k: handle.get(list(handle.keys())[0]).shape
for k, handle in handles.items()}
latent_dim = (TEST_ARCHITECTURE['composition']['neurons'][-1]
+ TEST_ARCHITECTURE['coverage']['neurons'][-1])
assert firsts['composition'] == (5, latent_dim)
assert firsts['coverage'] == (5, latent_dim)
fasta.unlink()
coverage.unlink()
for key, filename in output.items():
handles[key].close()
filename.unlink()
def test_load_model():
'''
Test if model can be loaded
'''
args = {'compo_neurons': TEST_ARCHITECTURE['composition']['neurons'],
'cover_neurons': TEST_ARCHITECTURE['coverage']['neurons'],
'cover_filters': TEST_ARCHITECTURE['coverage']['n_filters'],
'cover_kernel': TEST_ARCHITECTURE['coverage']['kernel_size'],
'cover_stride': TEST_ARCHITECTURE['coverage']['conv_stride'],
'merge_neurons': TEST_ARCHITECTURE['merge']['neurons'],
'kmer': 4, 'no_rc': True,
'fragment_length': FL, 'wsize': WSIZE, 'wstep': WSTEP}
cfg = Configuration()
cfg.init_config(output='.', **args)
cfg.io['h5'] = generate_h5_file(FL, filename='coverage.h5')
model = initialize_model('CoCoNet', cfg.get_input_shapes(), cfg.get_architecture())
model_path = Path('CoCoNet.pth')
torch.save({
'state': model.state_dict()
}, model_path)
loaded_model = load_model(cfg)
model_path.unlink()
cfg.io['h5'].unlink()
assert isinstance(loaded_model, CoCoNet)
def test_learn_save_load_model():
'''
Check:
- if the training goes through
- if the model is saved
'''
model = initialize_model('CoCoNet', TEST_SHAPES, TEST_ARCHITECTURE)
model_file = Path('{}/test_model.pth'.format(LOCAL_DIR))
results_file = Path('{}/test_res.csv'.format(LOCAL_DIR))
pair_files = {'train': Path('{}/pairs_train.npy'.format(LOCAL_DIR)),
'test': Path('{}/pairs_test.npy'.format(LOCAL_DIR))}
coverage_file = generate_h5_file(*TEST_CTG_LENGTHS, filename='coverage.h5')
fasta_file = generate_fasta_file(*TEST_CTG_LENGTHS, save=True)
fasta = [(seq.id, str(seq.seq)) for seq in SeqIO.parse(fasta_file, 'fasta')]
make_pairs(fasta, STEP, FL, output=pair_files['train'], n_examples=50)
make_pairs(fasta, STEP, FL, output=pair_files['test'], n_examples=5)
train(model, fasta_file, coverage_file, pair_files, results_file, output=model_file,
**TEST_LEARN_PRMS)
tests = model_file.is_file() and results_file.is_file()
for path in list(pair_files.values()) + [fasta_file, coverage_file, model_file, results_file]:
path.unlink()
assert tests
def test_init_combined_model():
'''
Check if model can be initialized
'''
model = initialize_model('CoCoNet', TEST_SHAPES, TEST_ARCHITECTURE)
assert isinstance(model, CoCoNet)
def test_init_coverage_model():
'''
Check if model can be initialized
'''
model = initialize_model('coverage',
TEST_SHAPES['coverage'],
TEST_ARCHITECTURE['coverage'])
assert isinstance(model, CoverageModel)
def test_init_composition_model():
'''
Check if model can be initialized
'''
model = initialize_model('composition',
TEST_SHAPES['composition'],
TEST_ARCHITECTURE['composition'])
assert isinstance(model, CompositionModel)
def test_coverage_model():
'''
Test if coverage model can compute an output
'''
model = initialize_model('coverage', TEST_SHAPES['coverage'], TEST_ARCHITECTURE['coverage'])
x_rd = [torch.FloatTensor(4, *TEST_SHAPES['coverage'][::-1]).random_(0, 10) for _ in range(2)]
y_rd = torch.FloatTensor(4, 1).random_(0, 1)
pred = model(*x_rd)
loss = model.compute_loss(pred, y_rd)
assert 'coverage' in pred
assert y_rd.shape == pred['coverage'].shape
assert isinstance(loss, torch.FloatTensor)
def generate_rd_model():
model = initialize_model('CoCoNet', TEST_SHAPES, TEST_ARCHITECTURE)
return model