Skip to content

Figure 8: Investigating root cause of G1 discrepancies: Kernel density estimation of the per-class packet size distributions.

import itertools

import numpy as np
import pandas as pd

import matplotlib as mpl
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib.colors import LogNorm, Normalize

%matplotlib inline
%config InlineBackend.figure_format='retina'

import tcbench as tcb
from tcbench import dataprep

FLOWPIC_DIM = 32
FLOWPIC_BLOCK_DURATION = 15

# load unfiltered dataset
dset = dataprep.FlowpicDataset(
    data=tcb.load_parquet(tcb.DATASETS.UCDAVISICDM19),
    timetofirst_colname="timetofirst",
    pkts_size_colname="pkts_size",
    pkts_dir_colname="pkts_dir",
    target_colname="app",
    flowpic_dim=FLOWPIC_DIM,
    flowpic_block_duration=FLOWPIC_BLOCK_DURATION,
)

REPLACE = {
    "google-doc": "G. Doc",
    "google-drive": "G. Drive",
    "google-music": "G. Music",
    "google-search": "G. Search",
    "youtube": "YouTube",
    "retraining-human-triggered": "Human",
    "retraining-script-triggered": "Script",
}

dset.df = dset.df.assign(
    app = dset.df["app"].replace(REPLACE),
    partition = dset.df["partition"].replace(REPLACE)
)

TARGETS_LABEL = sorted(dset.df["app"].unique())
PARTITIONS_NAME = sorted(dset.df["partition"].unique())

all_pkts_size = dict()

for partition_name in PARTITIONS_NAME:
    all_pkts_size[partition_name] = dict()

    for app in TARGETS_LABEL:
        df_tmp = dset.df[
            (dset.df["partition"] == partition_name) & (dset.df["app"] == app)
        ]

        l = []
        for idx in df_tmp.index:
            ser = df_tmp.loc[idx]
            indexes = np.where(ser["timetofirst"] < FLOWPIC_BLOCK_DURATION)[0]
            pkts_size = ser["pkts_size"][indexes]
            l.append(pkts_size)
        all_pkts_size[partition_name][app] = np.concatenate(l)

# WARNING: computing the KDE will take a few minutes

fig, axes = plt.subplots(nrows=1, ncols=5, figsize=(15, 5))

line_props = {
    "pretraining": dict(linestyle="-"),
    "Script": dict(
        linestyle=(0, (1, 1))
    ), 
    "Human": dict(linestyle=(0, (1, 1))),
}

for ax, app in zip(axes, TARGETS_LABEL):
    for partition_name in [
        "pretraining",
        "Script",
        "Human",
    ]:
        props = line_props[partition_name]
        sns.kdeplot(
            ax=ax,
            data=all_pkts_size[partition_name][app],
            linewidth=2,
            label=partition_name,
            **props,
            fill=True,
            alpha=0.1
        )
    ax.legend(bbox_to_anchor=(0.5, 1.5), loc="upper center")
    ax.set_title(app, fontsize=10)
    ax.set_xlim((-500, 1800))
    ax.set_xlabel("packet size")
    ax.set_ylabel("kde")

plt.tight_layout()
plt.savefig("ucdavid-icdm19_kde_pkts_size.png", dpi=300, bbox_inches='tight')

png