def test_encode_data_method():
m = 'random'
fp = 'data/train_data.csv'
nrows = 10
df = pd.read_csv(fp, sep='\t', nrows=nrows)
with pytest.raises(AttributeError):
encode_data(df, m)
def test_encode_data():
fp = 'data/train_data.csv'
nrows = 10
df = pd.read_csv(fp, sep='\t', nrows=nrows)
x, y = encode_data(df)
assert x.shape == (nrows, 280)
assert len(y) == nrows
def evaluate_model():
"""
Use this function to validate DeepMiRNA model
:return: keras history object with the results
"""
# get training parameters
model_name = gv.TRAIN_FINAL_MODEL_NAME
train_set_fp = os.path.join(gv.ROOT_DIR, gv.TRAIN_SET_LOCATION)
true_labels_fp = os.path.join(gv.ROOT_DIR, gv.TRUE_LABELS)
ohe_duplexes_fp = os.path.join(gv.ROOT_DIR, gv.ONE_HOT_ENCODED_DUPLEXES)
batch_size = gv.BATCH_SIZE
n_epochs = gv.N_EPOCHS
keep_prob = gv.KEEP_PROB
training_df = pd.read_csv(train_set_fp, sep='\t', usecols=gv.TRAIN_SET_COLUMNS)
# check if encoded data already exists
if Path(ohe_duplexes_fp).exists():
# load data
_logger.info(' One-hot encoded training set found. Loading data ...')
ytrain = np.loadtxt(true_labels_fp)
with h5py.File(ohe_duplexes_fp, 'r') as hf:
xtrain = hf['encoded_training_set'][:]
else :
_logger.info(' Encoding the training set. This might take some time')
xtrain, ytrain = encode_data(training_df)
# save for next computation
_logger.info(' Saving encoded data to disk.')
np.savetxt(true_labels_fp, ytrain)
with h5py.File(ohe_duplexes_fp, 'w') as hf:
hf.create_dataset('encoded_training_set', data=xtrain)
_logger.info(' Building and compiling the model')
model = _create_mlp_model(xtrain.shape[1], keep_prob)
_logger.info(' Training started')
history = train_eval(model, model_name, xtrain, ytrain, batch_size, n_epochs)
_plot_model_history(history)
_logger.info(' model {} saved.'.format(model_name))
_logger.info(' Best model achieved {:.2f} accuracy and {:.2f} validation loss on the validation set.'
.format(max(history.history['acc']), min(history.history['val_loss'])))
return history
def train_model():
"""
Train DeepMiRNA model over the whole training set and obtain the final model
:return: keras history object
"""
# get training parameters
model_name = gv.TRAIN_FINAL_MODEL_NAME
train_set_fp = os.path.join(gv.ROOT_DIR, gv.TRAIN_SET_LOCATION)
true_labels_fp = os.path.join(gv.ROOT_DIR, gv.TRUE_LABELS)
ohe_duplexes_fp = os.path.join(gv.ROOT_DIR, gv.ONE_HOT_ENCODED_DUPLEXES)
batch_size = gv.BATCH_SIZE
n_epochs = gv.N_EPOCHS
keep_prob = gv.KEEP_PROB
training_df = pd.read_csv(train_set_fp, sep='\t', usecols=gv.TRAIN_SET_COLUMNS)
# check if encoded data already exists
if Path(ohe_duplexes_fp).exists():
# load data
_logger.info(' One-hot encoded training set found. Loading data ...')
ytrain = np.loadtxt(true_labels_fp)
with h5py.File(ohe_duplexes_fp, 'r') as hf:
xtrain = hf['encoded_training_set'][:]
else :
_logger.info(' Encoding the training set. This might take some time')
xtrain, ytrain = encode_data(training_df)
# save for next computation
_logger.info(' Saving encoded data to disk.')
np.savetxt(true_labels_fp, ytrain)
with h5py.File(ohe_duplexes_fp, 'w') as hf:
hf.create_dataset('encoded_training_set', data=xtrain)
_logger.info(' Building and compiling the model')
model = _create_mlp_model(xtrain.shape[1], keep_prob)
_logger.info(' Training started')
history = train(model, model_name, xtrain, ytrain, batch_size, n_epochs)
return history
def test_encode_data_header():
d = {'a': [0], 'b': [1], 'c': [2]}
df = pd.DataFrame(d)
with pytest.raises(SystemExit):
encode_data(df)