import json import logging import os import numpy as np import pandas as pd import tensorflow as tf from keras.callbacks import CSVLogger, EarlyStopping, LambdaCallback, ModelCheckpoint from keras.models import Model, load_model import arguments import dataset import hyperband import models # create logger import visualize from arguments import get_model_args from utils import exists_or_make_path, get_custom_class_weights logger = logging.getLogger('logger') logger.setLevel(logging.DEBUG) # create console handler and set level to debug ch = logging.StreamHandler() ch.setLevel(logging.DEBUG) # create formatter formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s') # add formatter to ch ch.setFormatter(formatter) # add ch to logger logger.addHandler(ch) ch = logging.FileHandler("info.log") ch.setLevel(logging.DEBUG) # create formatter formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s') # add formatter to ch ch.setFormatter(formatter) # add ch to logger logger.addHandler(ch) args = arguments.parse() if args.gpu: config = tf.ConfigProto(log_device_placement=True) config.gpu_options.per_process_gpu_memory_fraction = 0.5 config.gpu_options.allow_growth = True session = tf.Session(config=config) # default parameter PARAMS = { "type": args.model_type, "depth": args.model_depth, # "batch_size": 64, "window_size": args.window, "domain_length": args.domain_length, "flow_features": 3, # 'dropout': 0.5, # currently fix 'domain_features': args.domain_embedding, 'embedding_size': args.embedding, 'flow_features': 3, 'filter_embedding': args.filter_embedding, 'dense_embedding': args.dense_embedding, 'kernel_embedding': args.kernel_embedding, 'filter_main': args.filter_main, 'dense_main': args.dense_main, 'kernel_main': args.kernel_main, 'input_length': 40, 'model_output': args.model_output } def create_model(model, output_type): if output_type == "both": return Model(inputs=[model.in_domains, model.in_flows], outputs=(model.out_client, model.out_server)) elif output_type == "client": return Model(inputs=[model.in_domains, model.in_flows], outputs=(model.out_client,)) else: raise Exception("unknown model output") def main_paul_best(): pauls_best_params = models.pauls_networks.best_config main_train(pauls_best_params) def main_hyperband(): params = { # static params "type": ["paul"], "batch_size": [args.batch_size], "window_size": [10], "domain_length": [40], "flow_features": [3], "input_length": [40], # model params "embedding_size": [8, 16, 32, 64, 128, 256], "filter_embedding": [8, 16, 32, 64, 128, 256], "kernel_embedding": [1, 3, 5, 7, 9], "hidden_embedding": [8, 16, 32, 64, 128, 256], "dropout": [0.5], "domain_features": [8, 16, 32, 64, 128, 256], "filter_main": [8, 16, 32, 64, 128, 256], "kernels_main": [1, 3, 5, 7, 9], "dense_main": [8, 16, 32, 64, 128, 256], } logger.info("create training dataset") domain_tr, flow_tr, name_tr, client_tr, server_tr = dataset.load_or_generate_h5data(args.train_h5data, args.train_data, args.domain_length, args.window) hp = hyperband.Hyperband(params, [domain_tr, flow_tr], [client_tr, server_tr]) results = hp.run() json.dump(results, open("hyperband.json")) def main_train(param=None): logger.info(f"Create model path {args.model_path}") exists_or_make_path(args.model_path) logger.info(f"Use command line arguments: {args}") domain_tr, flow_tr, name_tr, client_tr, server_windows_tr = dataset.load_or_generate_h5data(args.train_h5data, args.train_data, args.domain_length, args.window) logger.info("define callbacks") callbacks = [] callbacks.append(ModelCheckpoint(filepath=args.clf_model, monitor='loss', verbose=False, save_best_only=True)) callbacks.append(CSVLogger(args.train_log)) logger.info(f"Use early stopping: {args.stop_early}") if args.stop_early: callbacks.append(EarlyStopping(monitor='val_loss', patience=5, verbose=False)) custom_metrics = models.get_metric_functions() server_tr = np.max(server_windows_tr, axis=1) if args.class_weights: logger.info("class weights: compute custom weights") custom_class_weights = get_custom_class_weights(client_tr.value, server_tr) logger.info(custom_class_weights) else: logger.info("class weights: set default") custom_class_weights = None if not param: param = PARAMS logger.info(f"Generator model with params: {param}") embedding, model, new_model = models.get_models_by_params(param) callbacks.append(LambdaCallback( on_epoch_end=lambda epoch, logs: embedding.save(args.embedding_model)) ) model = create_model(model, args.model_output) new_model = create_model(new_model, args.model_output) if args.model_type in ("inter", "staggered"): server_tr = np.expand_dims(server_windows_tr, 2) model = new_model features = {"ipt_domains": domain_tr.value, "ipt_flows": flow_tr.value} if args.model_output == "both": labels = {"client": client_tr.value, "server": server_tr} loss_weights = {"client": 1.0, "server": 1.0} elif args.model_output == "client": labels = {"client": client_tr.value} loss_weights = {"client": 1.0} elif args.model_output == "server": labels = {"server": server_tr} loss_weights = {"server": 1.0} else: raise ValueError("unknown model output") logger.info(f"select model: {args.model_type}") if args.model_type == "staggered": logger.info("compile and pre-train server model") logger.info(model.get_config()) model.compile(optimizer='adam', loss='binary_crossentropy', loss_weights={"client": 0.0, "server": 1.0}, metrics=['accuracy'] + custom_metrics) model.fit(features, labels, batch_size=args.batch_size, epochs=args.epochs, class_weight=custom_class_weights) logger.info("fix server model") model.get_layer("domain_cnn").trainable = False model.get_layer("dense_server").trainable = False model.get_layer("server").trainable = False loss_weights = {"client": 1.0, "server": 0.0} logger.info("compile and train model") embedding.summary() logger.info(model.get_config()) model.compile(optimizer='adam', loss='binary_crossentropy', loss_weights=loss_weights, metrics=['accuracy'] + custom_metrics) model.summary() model.fit(features, labels, batch_size=args.batch_size, epochs=args.epochs, callbacks=callbacks, class_weight=custom_class_weights) def main_test(): logger.info("start test: load data") domain_val, flow_val, _, _, _, _ = dataset.load_or_generate_raw_h5data(args.test_h5data, args.test_data, args.domain_length, args.window) domain_encs, _ = dataset.load_or_generate_domains(args.test_data, args.domain_length) for model_args in get_model_args(args): results = {} logger.info(f"process model {model_args['model_path']}") clf_model = load_model(model_args["clf_model"], custom_objects=models.get_metrics()) pred = clf_model.predict([domain_val, flow_val], batch_size=args.batch_size, verbose=1) if args.model_output == "both": c_pred, s_pred = pred results["client_pred"] = c_pred results["server_pred"] = s_pred elif args.model_output == "client": results["client_pred"] = pred else: results["server_pred"] = pred embd_model = load_model(model_args["embedding_model"]) domain_embeddings = embd_model.predict(domain_encs, batch_size=args.batch_size, verbose=1) results["domain_embds"] = domain_embeddings dataset.save_predictions(model_args["model_path"], results) def main_visualization(): _, _, name_val, hits_vt, hits_trusted, server_val = dataset.load_or_generate_raw_h5data(args.test_h5data, args.test_data, args.domain_length, args.window) results = dataset.load_predictions(args.model_path) df = pd.DataFrame(data={ "names": name_val, "client_pred": results["client_pred"].flatten(), "hits_vt": hits_vt, "hits_trusted": hits_trusted }) df["client_val"] = np.logical_or(df.hits_vt == 1.0, df.hits_trusted >= 3) df_user = df.groupby(df.names).max() paul = dataset.load_predictions("results/paul/") df_paul = pd.DataFrame(data={ "names": paul["testNames"].flatten(), "client_pred": paul["testScores"].flatten(), "hits_vt": paul["testLabel"].flatten(), "hits_trusted": paul["testHits"].flatten() }) df_paul["client_val"] = np.logical_or(df_paul.hits_vt == 1.0, df_paul.hits_trusted >= 3) df_paul_user = df_paul.groupby(df_paul.names).max() logger.info("plot model") model = load_model(args.clf_model, custom_objects=models.get_metrics()) visualize.plot_model_as(model, os.path.join(args.model_path, "model.png")) # logger.info("plot training curve") # logs = pd.read_csv(args.train_log) # if "acc" in logs.keys(): # visualize.plot_training_curve(logs, "", "{}/client_train.png".format(args.model_path)) # elif "client_acc" in logs.keys() and "server_acc" in logs.keys(): # visualize.plot_training_curve(logs, "client_", "{}/client_train.png".format(args.model_path)) # visualize.plot_training_curve(logs, "server_", "{}/server_train.png".format(args.model_path)) # else: # logger.warning("Error while plotting training curves") logger.info("plot pr curve") visualize.plot_clf() visualize.plot_precision_recall(df.client_val.as_matrix(), df.client_pred.as_matrix(), args.model_name) visualize.plot_precision_recall(df_paul.client_val.as_matrix(), df_paul.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save("{}/window_client_prc.png".format(args.model_path)) logger.info("plot roc curve") visualize.plot_clf() visualize.plot_roc_curve(df.client_val.as_matrix(), df.client_pred.as_matrix(), args.model_name) visualize.plot_roc_curve(df_paul.client_val.as_matrix(), df_paul.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save("{}/window_client_roc.png".format(args.model_path)) visualize.plot_clf() visualize.plot_precision_recall(df_user.client_val.as_matrix(), df_user.client_pred.as_matrix(), args.model_name) visualize.plot_precision_recall(df_paul_user.client_val.as_matrix(), df_paul_user.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save("{}/user_client_prc.png".format(args.model_path)) visualize.plot_clf() visualize.plot_roc_curve(df_user.client_val.as_matrix(), df_user.client_pred.as_matrix(), args.model_name) visualize.plot_roc_curve(df_paul_user.client_val.as_matrix(), df_paul_user.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save("{}/user_client_roc.png".format(args.model_path)) # absolute values visualize.plot_confusion_matrix(df.client_val.as_matrix(), df.client_pred.as_matrix().round(), "{}/client_cov.png".format(args.model_path), normalize=False, title="Client Confusion Matrix") visualize.plot_confusion_matrix(df_user.client_val.as_matrix(), df_user.client_pred.as_matrix().round(), "{}/user_cov.png".format(args.model_path), normalize=False, title="User Confusion Matrix") # normalized visualize.plot_confusion_matrix(df.client_val.as_matrix(), df.client_pred.as_matrix().round(), "{}/client_cov_norm.png".format(args.model_path), normalize=True, title="Client Confusion Matrix") visualize.plot_confusion_matrix(df_user.client_val.as_matrix(), df_user.client_pred.as_matrix().round(), "{}/user_cov_norm.png".format(args.model_path), normalize=True, title="User Confusion Matrix") 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)) def main_visualize_all(): _, _, name_val, hits_vt, hits_trusted, server_val = dataset.load_or_generate_raw_h5data(args.test_h5data, args.test_data, args.domain_length, args.window) def load_df(path): res = dataset.load_predictions(path) res = pd.DataFrame(data={ "names": name_val, "client_pred": res["client_pred"].flatten(), "hits_vt": hits_vt, "hits_trusted": hits_trusted }) res["client_val"] = np.logical_or(res.hits_vt == 1.0, res.hits_trusted >= 3) return res paul = dataset.load_predictions("results/paul/") df_paul = pd.DataFrame(data={ "names": paul["testNames"].flatten(), "client_pred": paul["testScores"].flatten(), "hits_vt": paul["testLabel"].flatten(), "hits_trusted": paul["testHits"].flatten() }) df_paul["client_val"] = np.logical_or(df_paul.hits_vt == 1.0, df_paul.hits_trusted >= 3) df_paul_user = df_paul.groupby(df_paul.names).max() logger.info("plot pr curves") visualize.plot_clf() for model_args in get_model_args(args): df = load_df(model_args["model_path"]) visualize.plot_precision_recall(df.client_val.as_matrix(), df.client_pred.as_matrix(), model_args["model_name"]) visualize.plot_precision_recall(df_paul.client_val.as_matrix(), df_paul.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save(f"{args.output_prefix}_window_client_prc.png") logger.info("plot roc curves") visualize.plot_clf() for model_args in get_model_args(args): df = load_df(model_args["model_path"]) visualize.plot_roc_curve(df.client_val.as_matrix(), df.client_pred.as_matrix(), model_args["model_name"]) visualize.plot_roc_curve(df_paul.client_val.as_matrix(), df_paul.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save(f"{args.output_prefix}_window_client_roc.png") logger.info("plot user pr curves") visualize.plot_clf() for model_args in get_model_args(args): df = load_df(model_args["model_path"]) df = df.groupby(df.names).max() visualize.plot_precision_recall(df.client_val.as_matrix(), df.client_pred.as_matrix(), model_args["model_name"]) visualize.plot_precision_recall(df_paul_user.client_val.as_matrix(), df_paul_user.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save(f"{args.output_prefix}_user_client_prc.png") logger.info("plot user roc curves") visualize.plot_clf() for model_args in get_model_args(args): df = load_df(model_args["model_path"]) df = df.groupby(df.names).max() visualize.plot_roc_curve(df.client_val.as_matrix(), df.client_pred.as_matrix(), model_args["model_name"]) visualize.plot_roc_curve(df_paul_user.client_val.as_matrix(), df_paul_user.client_pred.as_matrix(), "paul") visualize.plot_legend() visualize.plot_save(f"{args.output_prefix}_user_client_roc.png") def main(): if "train" == args.mode: main_train() if "hyperband" == args.mode: main_hyperband() if "test" == args.mode: main_test() if "fancy" == args.mode: main_visualization() if "all_fancy" == args.mode: main_visualize_all() if "paul" == args.mode: main_paul_best() if __name__ == "__main__": main()