add tsne (does not work with big data)

fix model loading with custom selu function

main.py

@@ -248,7 +248,7 @@ def main_test():
         else:
             results["server_pred"] = pred
 
-        embd_model = load_model(model_args["embedding_model"])
+        embd_model = load_model(model_args["embedding_model"], custom_objects=models.get_metrics())
         domain_embeddings = embd_model.predict(domain_encs, batch_size=args.batch_size, verbose=1)
         results["domain_embds"] = domain_embeddings
 
@@ -333,7 +333,17 @@ def main_visualization():
     logger.info("visualize embedding")
     domain_encs, labels = dataset.load_or_generate_domains(args.test_data, args.domain_length)
     domain_embedding = results["domain_embds"]
-    visualize.plot_embedding(domain_embedding, labels, path="{}/embd.png".format(args.model_path))
+    visualize.plot_embedding(domain_embedding, labels, path="{}/embd_svd.png".format(args.model_path), method="svd")
+    visualize.plot_embedding(domain_embedding, labels, path="{}/embd_tsne.png".format(args.model_path), method="tsne")
+
+
+def plot_embedding():
+    logger.info("visualize embedding")
+    results = dataset.load_predictions(args.model_path)
+    domain_encs, labels = dataset.load_or_generate_domains(args.test_data, args.domain_length)
+    domain_embedding = results["domain_embds"]
+    visualize.plot_embedding(domain_embedding, labels, path="{}/embd_svd.png".format(args.model_path), method="svd")
+    visualize.plot_embedding(domain_embedding, labels, path="{}/embd_tsne.png".format(args.model_path), method="tsne")
 
 
 def main_visualize_all():
@@ -409,6 +419,8 @@ def main():
         main_visualization()
     if "all_fancy" == args.mode:
         main_visualize_all()
+    if "embd" == args.mode:
+        plot_embedding()
     if "paul" == args.mode:
         main_paul_best()
 

models/__init__.py

@@ -1,5 +1,6 @@
 import keras.backend as K
 
+from models.renes_networks import selu
 from . import flat_2, pauls_networks, renes_networks
 
 
@@ -24,11 +25,11 @@ def get_models_by_params(params: dict):
     dense_dim = params.get("dense_main")
     model_output = params.get("model_output", "both")
     # create models
-    if network_depth == "small":
+    if network_depth == "flat1":
         networks = pauls_networks
-    elif network_depth == "flat":
+    elif network_depth == "flat2":
         networks = flat_2
-    elif network_depth == "medium":
+    elif network_depth == "deep1":
         networks = renes_networks
     else:
         raise Exception("network not found")
@@ -49,6 +50,7 @@ def get_metrics():
         ("precision", precision),
         ("recall", recall),
         ("f1_score", f1_score),
+        ("selu", selu)
     ])
 
 

models/renes_networks.py

@@ -32,7 +32,7 @@ def get_embedding(embedding_size, input_length, filter_size, kernel_size, hidden
     y = Conv1D(filter_size, kernel_size=3, activation=selu)(y)
     y = Conv1D(filter_size, kernel_size=3, activation=selu)(y)
     y = GlobalAveragePooling1D()(y)
-    y = Dense(hidden_dims, activation="relu")(y)
+    y = Dense(hidden_dims, activation=selu)(y)
     return KerasModel(x, y)
 
 
@@ -53,7 +53,7 @@ def get_model(cnnDropout, flow_features, domain_features, window_size, domain_le
     y = GlobalMaxPooling1D()(y)
     y = Dropout(cnnDropout)(y)
     y = Dense(dense_dim, activation=selu)(y)
-    y = Dense(dense_dim // 2, activation=selu)(y)
+    y = Dense(dense_dim, activation=selu)(y)
     out_client = Dense(1, activation='sigmoid', name="client")(y)
     out_server = Dense(1, activation='sigmoid', name="server")(y)
 
@@ -67,6 +67,9 @@ def get_new_model(dropout, flow_features, domain_features, window_size, domain_l
     encoded = TimeDistributed(cnn, name="domain_cnn")(ipt_domains)
     merged = keras.layers.concatenate([encoded, ipt_flows], -1)
     y = Dense(dense_dim, activation=selu)(merged)
+    y = Dense(dense_dim,
+              activation="relu",
+              name="dense_server")(y)
     out_server = Dense(1, activation="sigmoid", name="server")(y)
     merged = keras.layers.concatenate([merged, y], -1)
     # CNN processing a small slides of flow windows
@@ -90,6 +93,7 @@ def get_new_model(dropout, flow_features, domain_features, window_size, domain_l
     # remove temporal dimension by global max pooling
     y = GlobalMaxPooling1D()(y)
     y = Dropout(dropout)(y)
+    y = Dense(dense_dim, activation=selu)(y)
     y = Dense(dense_dim,
               activation=selu,
               name="dense_client")(y)

visualize.py

@@ -3,6 +3,7 @@ import os
 import matplotlib.pyplot as plt
 import numpy as np
 from sklearn.decomposition import TruncatedSVD
+from sklearn.manifold import TSNE
 from sklearn.metrics import (
     auc, classification_report, confusion_matrix, fbeta_score, precision_recall_curve,
     roc_auc_score, roc_curve
@@ -155,10 +156,13 @@ def plot_training_curve(logs, key, path, dpi=600):
     plt.close()
 
 
-def plot_embedding(domain_embedding, labels, path, dpi=600):
+def plot_embedding(domain_embedding, labels, path, dpi=600, method="svd"):
-    svd = TruncatedSVD(n_components=2)
+    if method == "svd":
-    domain_reduced = svd.fit_transform(domain_embedding)
+        red = TruncatedSVD(n_components=2)
-    print(svd.explained_variance_ratio_)
+    elif method == "tsne":
+        red = TSNE(n_components=2, verbose=2)
+    domain_reduced = red.fit_transform(domain_embedding)
+    print(red.explained_variance_ratio_)
     # use if draw subset of predictions
     # idx = np.random.choice(np.arange(len(domain_reduced)), 10000)
     plt.scatter(domain_reduced[:, 0],