def impute_from_model(self,
models_folder: str,
matrices: iter,
postprocess=True):
"""Generator to provide imputed matrices on-the-fly
Arguments:
models_folder {str} -- Path to directory containing trained CpGNet models
matrices {iter} -- An iterable containging n x m matrices with n=cpgs and m=reads
Keyword Arguments:
postprocess {bool} -- Round imputed values to 1s and 0s (default: {True})
"""
model_path = os.path.join(
models_folder,
"saved_model_{}_cpgs.prelim".format(self.cpg_density))
trained_model = PReLIM(cpgDensity=self.cpg_density)
print("Successfully loaded model: {}".format(model_path), flush=True)
trained_model.model = load(model_path)
for m in matrices:
# only impute if there is an unknown
if -1 in m:
m = m.astype(float)
pm = trained_model.impute(m)
if postprocess:
pm = self.postprocess_predictions(pm)
# Nothing to impute, passback original matrix to keep list in order
else:
pm = m.copy()
# K.clear_session()
yield pm
def setUp(self):
self.required_data = [bamA, prelim_model]
check_data_exists(self.required_data)
parser = ParseBam.BamFileReadParser(
os.path.join(test_data_location, bamA), 20)
reads = parser.parse_reads("chr1", 910600, 910700)
self.matrix = parser.create_matrix(reads).dropna(how="all")
self.imputer = Imputation(4, os.path.join(test_data_location, bamA))
self.prelim = PReLIM(4)
self.prelim.model = load(os.path.join(test_data_location,
prelim_model))
self.predictions = self.prelim.impute(np.array(self.matrix.fillna(-1)))
self.imputed_matrix = self.imputer.postprocess_predictions(
self.predictions)
self.imputed_matrix = pd.DataFrame(self.imputed_matrix).dropna()
def __init__(self, cpg_density=None, save_path=None):
"""
Class to train a CpGNet model from input data
:param cpg_density: Number of CpGs
:type cpg_density: int
:param save_path: Location of folder to save the resulting model files. One per cpg density
"""
if not cpg_density:
raise AttributeError("CpG density must be specified")
if not save_path:
raise AttributeError(
"Folder to save trained model must be specified")
self.save_path = save_path
self.cpg_density = cpg_density
self.model = PReLIM(cpgDensity=cpg_density)
class TrainWithPReLIM:
"""
Used to train models using CpGnet
"""
def __init__(self, cpg_density=None, save_path=None):
"""
Class to train a CpGNet model from input data
:param cpg_density: Number of CpGs
:type cpg_density: int
:param save_path: Location of folder to save the resulting model files. One per cpg density
"""
if not cpg_density:
raise AttributeError("CpG density must be specified")
if not save_path:
raise AttributeError(
"Folder to save trained model must be specified")
self.save_path = save_path
self.cpg_density = cpg_density
self.model = PReLIM(cpgDensity=cpg_density)
def save_net(self, model):
"""
Save the network to a file
:param model: The trained PReLIM model. Located at PReLIM.model
:type model: :class:`clubcpg_prelim.PReLIM`
:return: Path to the saved model
"""
file_name = "saved_model_{}_cpgs.prelim".format(self.cpg_density)
output = os.path.join(self.save_path, file_name)
dump(model, output)
print("Saved {} cpg model to {}".format(self.cpg_density, output))
return output
def train_model(self, bins: iter):
"""
Train the CpGNet model on a list of provided bins
:param bins: iterable containing CpG matrices of 1 (methylated), 0 (unmethylated), and -1 (unknown)
:return: Path to the saved model file
"""
self.model.train(bins, model_file="no")
output = self.save_net(self.model.model)
return output