replace softmax by sigmoid in final layer, also adjust dataset for that

dataset.py

@@ -6,7 +6,6 @@ from multiprocessing import Pool
 import h5py
 import numpy as np
 import pandas as pd
-from keras.utils import np_utils
 from tqdm import tqdm
 
 logger = logging.getLogger('logger')
@@ -119,7 +118,7 @@ def create_dataset_from_flows(user_flow_df, char_dict, max_len, window_size=10):
     client_tr[:pos_idx.shape[-1]] = 1.0
     server_tr = server_tr[idx]
 
-    client_tr = np_utils.to_categorical(client_tr, 2)
+    # client_tr = np_utils.to_categorical(client_tr, 2)
 
     return domain_tr, flow_tr, client_tr, server_tr
 

main.py

@@ -7,7 +7,6 @@ import pandas as pd
 import tensorflow as tf
 from keras.callbacks import ModelCheckpoint, CSVLogger, EarlyStopping
 from keras.models import load_model
-from keras.utils import np_utils
 from sklearn.utils import class_weight
 
 import arguments
@@ -98,8 +97,8 @@ def main_hyperband():
 
 
 def get_custom_class_weights(client_tr, server_tr):
-    client = client_tr.value.argmax(1) if type(client_tr) != np.ndarray else client_tr.argmax(1)
+    client = client_tr.value if type(client_tr) != np.ndarray else client_tr
-    server = server_tr.value.argmax(1) if type(server_tr) != np.ndarray else server_tr.argmax(1)
+    server = server_tr.value if type(server_tr) != np.ndarray else server_tr
     client_class_weight = class_weight.compute_class_weight('balanced', np.unique(client), client)
     server_class_weight = class_weight.compute_class_weight('balanced', np.unique(server), server)
     return {
@@ -157,10 +156,10 @@ def main_train(param=None):
     logger.info("compile model")
     custom_metrics = models.get_metric_functions()
     model.compile(optimizer='adam',
-                  loss='categorical_crossentropy',
+                  loss='binary_crossentropy',
                   metrics=['accuracy'] + custom_metrics)
 
-    server_tr = np_utils.to_categorical(np.max(server_windows_tr, axis=1), 2)
+    server_tr = np.max(server_windows_tr, axis=1)
 
     if args.class_weights:
         logger.info("class weights: compute custom weights")
@@ -261,10 +260,10 @@ def main_new_model():
     logger.info("compile model")
     custom_metrics = models.get_metric_functions()
     model.compile(optimizer='adam',
-                  loss='categorical_crossentropy',
+                  loss='binary_crossentropy',
                   metrics=['accuracy'] + custom_metrics)
 
-    server_tr = np_utils.to_categorical(np.max(server_windows_tr, axis=1), 2)
+    server_tr = np.max(server_windows_tr, axis=1)
 
     if args.class_weights:
         logger.info("class weights: compute custom weights")
@@ -274,7 +273,7 @@ def main_new_model():
         logger.info("class weights: set default")
         custom_class_weights = None
     logger.info("start training")
-    server_tr = np.stack(np_utils.to_categorical(s, 2) for s in server_windows_tr)
+    server_tr = np.expand_dims(server_windows_tr, 2)
     model.fit([domain_tr, flow_tr],
               [client_tr, server_tr],
               batch_size=args.batch_size,

models/pauls_networks.py

@@ -43,7 +43,6 @@ def get_model(cnnDropout, flow_features, domain_features, window_size, domain_le
     ipt_flows = Input(shape=(window_size, flow_features), name="ipt_flows")
     merged = keras.layers.concatenate([encoded, ipt_flows], -1)
     # CNN processing a small slides of flow windows
-    # TODO: add more layers?
     y = Conv1D(cnn_dims,
                kernel_size,
                activation='relu',
@@ -52,8 +51,8 @@ def get_model(cnnDropout, flow_features, domain_features, window_size, domain_le
     y = GlobalMaxPooling1D()(y)
     y = Dropout(cnnDropout)(y)
     y = Dense(dense_dim, activation='relu')(y)
-    y1 = Dense(2, activation='softmax', name="client")(y)
+    y1 = Dense(1, activation='sigmoid', name="client")(y)
-    y2 = Dense(2, activation='softmax', name="server")(y)
+    y2 = Dense(1, activation='sigmoid', name="server")(y)
 
     return Model(inputs=[ipt_domains, ipt_flows], outputs=(y1, y2))
 
@@ -64,7 +63,7 @@ def get_new_model(dropout, flow_features, domain_features, window_size, domain_l
     ipt_flows = Input(shape=(window_size, flow_features), name="ipt_flows")
     encoded = TimeDistributed(cnn)(ipt_domains)
 
-    y2 = Dense(2, activation="softmax", name="server")(encoded)
+    y2 = Dense(1, activation="sigmoid", name="server")(encoded)
     merged = keras.layers.concatenate([encoded, ipt_flows, y2], -1)
 
     y = Conv1D(cnn_dims,
@@ -76,7 +75,7 @@ def get_new_model(dropout, flow_features, domain_features, window_size, domain_l
     y = Dropout(dropout)(y)
     y = Dense(dense_dim, activation='relu')(y)
 
-    y1 = Dense(2, activation='softmax', name="client")(y)
+    y1 = Dense(1, activation='sigmoid', name="client")(y)
     model = Model(inputs=[ipt_domains, ipt_flows], outputs=(y1, y2))
 
     return model

models/renes_networks.py

@@ -9,13 +9,9 @@ def get_embedding(vocab_size, embedding_size, input_length,
     x = y = Input(shape=(input_length,))
     y = Embedding(input_dim=vocab_size, output_dim=embedding_size)(y)
     y = Conv1D(filter_size, kernel_size=5, activation='relu')(y)
-    # NOTE: max pooling destroys information flow for embedding
-    # y = MaxPool1D(pool_size=3, strides=1)(y)
     y = Conv1D(filter_size, kernel_size=3, activation='relu')(y)
-    # y = MaxPool1D(pool_size=3, strides=1)(y)
     y = Conv1D(filter_size, kernel_size=3, activation='relu')(y)
     y = GlobalAveragePooling1D()(y)
-    # y = Dropout(drop_out)(y)
     y = Dense(hidden_dims, activation="relu")(y)
     return Model(x, y)
 
@@ -38,8 +34,8 @@ def get_model(cnnDropout, flow_features, domain_features, window_size, domain_le
     y = Dropout(cnnDropout)(y)
     y = Dense(dense_dim, activation='relu')(y)
     y = Dense(dense_dim // 2, activation='relu')(y)
-    y1 = Dense(2, activation='softmax', name="client")(y)
+    y1 = Dense(1, activation='sigmoid', name="client")(y)
-    y2 = Dense(2, activation='softmax', name="server")(y)
+    y2 = Dense(1, activation='sigmoid', name="server")(y)
 
     return Model(inputs=[ipt_domains, ipt_flows], outputs=(y1, y2))
 
@@ -50,7 +46,7 @@ def get_new_model(dropout, flow_features, domain_features, window_size, domain_l
     ipt_flows = Input(shape=(window_size, flow_features), name="ipt_flows")
     encoded = TimeDistributed(cnn)(ipt_domains)
 
-    y2 = Dense(2, activation="softmax", name="server")(encoded)
+    y2 = Dense(1, activation="sigmoid", name="server")(encoded)
     merged = keras.layers.concatenate([encoded, ipt_flows, y2], -1)
 
     y = Conv1D(cnn_dims,
@@ -62,7 +58,7 @@ def get_new_model(dropout, flow_features, domain_features, window_size, domain_l
     y = Dropout(dropout)(y)
     y = Dense(dense_dim, activation='relu')(y)
 
-    y1 = Dense(2, activation='softmax', name="client")(y)
+    y1 = Dense(1, activation='sigmoid', name="client")(y)
     model = Model(inputs=[ipt_domains, ipt_flows], outputs=(y1, y2))
 
     return model