diff --git a/ecg/featurize.py b/ecg/featurize.py
index a7e68ee3..0f128908 100644
--- a/ecg/featurize.py
+++ b/ecg/featurize.py
@@ -5,7 +5,6 @@
 import warnings
 import scipy.signal as scs
 
-
 class Normalizer(object):
     def __init__(self, strategy):
         self.scaler = None
diff --git a/ecg/network.py b/ecg/network.py
index 8062e3b1..c009ca12 100644
--- a/ecg/network.py
+++ b/ecg/network.py
@@ -1,6 +1,5 @@
 from keras import backend as K
 
-
 def _bn_relu(layer, dropout=0, **params):
     from keras.layers import BatchNormalization
     from keras.layers import Activation
@@ -13,7 +12,6 @@ def _bn_relu(layer, dropout=0, **params):
 
     return layer
 
-
 def add_conv_weight(
         layer,
         filter_length,
@@ -41,7 +39,6 @@ def add_conv_layers(layer, **params):
         layer = _bn_relu(layer, **params)
     return layer
 
-
 def resnet_block(
         layer,
         num_filters,
@@ -83,12 +80,10 @@ def zeropad_output_shape(input_shape):
     layer = merge([shortcut, layer], mode="sum")
     return layer
 
-
 def get_num_filters_at_index(index, num_start_filters, **params):
     return 2**int(index / params["conv_increase_channels_at"]) \
         * num_start_filters
 
-
 def add_resnet_layers(layer, **params):
     layer = add_conv_weight(
         layer,
@@ -109,14 +104,12 @@ def add_resnet_layers(layer, **params):
     layer = _bn_relu(layer, **params)
     return layer
 
-
 def add_output_layer(layer, **params):
     from keras.layers.core import Dense, Activation
     from keras.layers.wrappers import TimeDistributed
     layer = TimeDistributed(Dense(params["num_categories"]))(layer)
     return Activation('softmax')(layer)
 
-
 def add_compile(model, **params):
     if params["optimizer"] == "adam":
         from keras.optimizers import Adam
@@ -136,7 +129,6 @@ def add_compile(model, **params):
                   optimizer=optimizer,
                   metrics=['accuracy'])
 
-
 def build_network(**params):
     from keras.models import Model
     from keras.layers import Input
diff --git a/ecg/predict.py b/ecg/predict.py
index 9c6e574a..b0a34161 100644
--- a/ecg/predict.py
+++ b/ecg/predict.py
@@ -1,24 +1,18 @@
-from builtins import str
 import argparse
 import numpy as np
 import json
 import os
 import load
 import util
-from joblib import Memory
 import time
 import pickle
 
-memory = Memory(cachedir='./cache')
-
-@memory.cache
 def get_model_pred_probs(model_path, x):
     from keras.models import load_model
     model = load_model(model_path)
     probs = model.predict(x, verbose=1)
     return probs
 
-
 def get_ensemble_pred_probs(model_paths, x):
     print("Averaging " + str(len(model_paths)) + " model predictions...")
     all_model_probs = [get_model_pred_probs(model_path, x)
@@ -38,7 +32,6 @@ def get_folder_name(start_time):
         os.makedirs(folder_name)
     return folder_name
 
-
 def predict(args, train_params, test_params):
     x, gt, processor, _ = load.load_test(
         test_params,
@@ -56,14 +49,12 @@ def predict(args, train_params, test_params):
 
     save_predictions(folder_name, x, gt, probs, processor)
 
-
 def save_predictions(path, x, gt, probs, processor):
     np.save(path + '/x', x)
     np.save(path + '/gt', gt)
     np.save(path + '/probs', probs)
     pickle.dump(processor, open(path + '/processor', 'w+'))
 
-
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
     parser.add_argument("test_config_file", help="path to config file")
diff --git a/ecg/train.py b/ecg/train.py
index c114d79d..eede36bf 100644
--- a/ecg/train.py
+++ b/ecg/train.py
@@ -5,6 +5,7 @@
 from builtins import open
 from builtins import int
 from builtins import str
+
 import argparse
 import json
 import os
@@ -17,20 +18,17 @@
 
 MAX_EPOCHS = 100
 
-
 def get_folder_name(start_time, experiment_name):
     folder_name = FOLDER_TO_SAVE + experiment_name + '/' + start_time
     if not os.path.exists(folder_name):
         os.makedirs(folder_name)
     return folder_name
 
-
 def get_filename_for_saving(start_time, experiment_name):
     saved_filename = get_folder_name(start_time, experiment_name) + \
         "/{val_loss:.3f}-{val_acc:.3f}-{epoch:03d}-{loss:.3f}-{acc:.3f}.hdf5"
     return saved_filename
 
-
 def plot_model(model, start_time, experiment_name):
     from keras.utils.visualize_util import plot
     plot(
@@ -39,7 +37,6 @@ def plot_model(model, start_time, experiment_name):
         show_shapes=True,
         show_layer_names=False)
 
-
 def save_params(params, start_time, experiment_name):
     saving_filename = get_folder_name(start_time, experiment_name) + \
         "/params.json"
@@ -49,20 +46,17 @@ def save_params(params, start_time, experiment_name):
     with open(saving_filename, 'w') as outfile:
         outfile.write(save_str)
 
-
 def get_augment_fn(params):
     if params.get("amplitude_scale", False):
         return amplitude_scale
     else:
         return None
 
-
 def amplitude_scale(x):
     scales = np.random.uniform(low=0.25, high=2.0,
                                size=(x.shape[0], 1, 1))
     return scales * x
 
-
 def data_generator(x, y, batch_size, augmenter=None):
     num_examples = x.shape[0]
     indices = np.arange(num_examples)
@@ -77,7 +71,6 @@ def data_generator(x, y, batch_size, augmenter=None):
                 x_mb = augmenter(x_mb)
             yield (x_mb, y_mb)
 
-
 def train(args, params):
     global FOLDER_TO_SAVE
 
@@ -88,7 +81,7 @@ def train(args, params):
             if "dropout" in key:
                 params[key] = 0
 
-    params["test_split_start"] = args.test_split_start        
+    params["test_split_start"] = args.test_split_start
 
     dl, processor = load.load_train(params)
 
@@ -162,7 +155,6 @@ def train(args, params):
         callbacks=[checkpointer, reduce_lr, stopping],
         verbose=args.verbose)
 
-
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
     parser.add_argument("config_file", help="path to config file")