shithub: opus

Info  •  Files  •  Log  •  Branches

Using lost packet file instead of uniform random

--- a/dnn/training_tf2/plc_loader.py

+++ b/dnn/training_tf2/plc_loader.py

@@ -29,19 +29,25 @@

 from tensorflow.keras.utils import Sequence

 class PLCLoader(Sequence):

-    def __init__(self, features, batch_size):

+    def __init__(self, features, lost, batch_size):

         self.batch_size = batch_size

         self.nb_batches = features.shape[0]//self.batch_size

-        self.features = features[:self.nb_batches*self.batch_size, :]

+        self.features = features[:self.nb_batches*self.batch_size, :, :]

+        self.lost = lost.astype('float')

+        self.lost = self.lost[:(len(self.lost)//features.shape[1]-1)*features.shape[1]]

         self.on_epoch_end()

     def on_epoch_end(self):

         self.indices = np.arange(self.nb_batches*self.batch_size)

         np.random.shuffle(self.indices)

+        offset = np.random.randint(0, high=self.features.shape[1])

+        self.lost_offset = np.reshape(self.lost[offset:-self.features.shape[1]+offset], (-1, self.features.shape[1]))

+        self.lost_indices = np.random.randint(0, high=self.lost_offset.shape[0], size=self.nb_batches*self.batch_size)

     def __getitem__(self, index):

         features = self.features[self.indices[index*self.batch_size:(index+1)*self.batch_size], :, :]

-        lost = (np.random.rand(features.shape[0], features.shape[1]) > .2).astype('float')

+        #lost = (np.random.rand(features.shape[0], features.shape[1]) > .2).astype('float')

+        lost = self.lost_offset[self.lost_indices[index*self.batch_size:(index+1)*self.batch_size], :]

         lost = np.reshape(lost, (features.shape[0], features.shape[1], 1))

         lost_mask = np.tile(lost, (1,1,features.shape[2]))

--- a/dnn/training_tf2/train_plc.py

+++ b/dnn/training_tf2/train_plc.py

@@ -34,6 +34,7 @@

 parser = argparse.ArgumentParser(description='Train a PLC model')

 parser.add_argument('features', metavar='<features file>', help='binary features file (float32)')

+parser.add_argument('lost_file', metavar='<packet loss file>', help='packet loss traces (int8)')

 parser.add_argument('output', metavar='<output>', help='trained model file (.h5)')

 parser.add_argument('--model', metavar='<model>', default='lpcnet_plc', help='PLC model python definition (without .py)')

 group1 = parser.add_mutually_exclusive_group()

@@ -151,6 +152,7 @@

 features = features[:, :, :nb_used_features]

+lost = np.memmap(args.lost_file, dtype='int8', mode='r')

 # dump models to disk as we go

 checkpoint = ModelCheckpoint('{}_{}_{}.h5'.format(args.output, args.gru_size, '{epoch:02d}'))

@@ -164,7 +166,7 @@

 model.save_weights('{}_{}_initial.h5'.format(args.output, args.gru_size))

-loader = PLCLoader(features, batch_size)

+loader = PLCLoader(features, lost, batch_size)

 callbacks = [checkpoint]

 if args.logdir is not None:

-- 

⑨