Network flow visualization used by Dynatrace - SAL - LIT.AI.JKU in the NAD 2021 challenge

network_flow_visualization.py

# Copyright 2021
#    Dynatrace Research
#    SAL Silicon Austria Labs
#    LIT Artificial Intelligence Lab
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from typing import List, Dict

import pandas as pd

import datashader as ds
import holoviews as hv
import holoviews.operation.datashader as hd

from IPython.core.display import display, HTML

hv.extension("bokeh")


def _get_data_matrix(df: pd.DataFrame, groups: List[str]) -> pd.DataFrame:
    # merge those group columns into a group label and get the group x time grouping
    df["group"] = df.groupby(groups).ngroup()
    mat = df.groupby(["group", "time"])["label"].agg(["count", "median"]).astype(int)

    # sort by group size and re-index groups
    num_groups = mat.index.get_level_values(0).nunique()

    # add the broadcasted group size to the "group" level
    sizes = df.groupby("group").size()
    _, sizes = mat.align(sizes, level=0, axis=0)
    mat["size"] = sizes

    # sort by the group size, then by group ids
    mat = mat.reset_index().sort_values(by=["size", "group"])

    # modify the group ids so that they are ascending again
    mapping = dict(zip(mat["group"].unique(), range(num_groups)))
    mat["group"] = mat["group"].map(mapping)

    return mat


def shade_network_flow(df: pd.DataFrame, groups: List[str], colors: Dict[int, str], legend: bool = True):
    if legend:
        html = "<br/>".join([
            f"<span style=\"background-color:black;color:{val};font-weight:bold;\">" +
            f"{key} : {val}" +
            f"</span>"
            for key, val in colors.items()
        ])
        display(HTML(html))

    matrix = _get_data_matrix(df, groups)
    points = hv.Points(matrix, ["group", "time"])
    shaded = hd.datashade(points, aggregator=ds.count_cat("median"), color_key=colors)
    render = hd.dynspread(shaded, threshold=0.5, max_px=4).opts(
        bgcolor="black", xaxis=None, yaxis=None, width=1200, height=600)

    return render

network_flow_visualization_nb.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "saving-olive",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from network_flow_visualization import shade_network_flow "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "conventional-mason",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_synthetic_data(size: int = 1_000_000, n_classes: int = 5) -> pd.DataFrame:\n",
    "    assert size >= 1_000\n",
    "    assert n_classes >= 2\n",
    "\n",
    "    # sample timestamps from the current day\n",
    "    today = np.datetime64(\"today\", \"s\")\n",
    "    tomorrow = today + np.timedelta64(1, \"D\")\n",
    "    timestamps = np.random.randint(today.astype(int), tomorrow.astype(int), size=size)\n",
    "    timestamps = np.sort(timestamps)\n",
    "\n",
    "    # sample some labels with a strong bias towards the first class\n",
    "    weights = np.array([2 * n_classes] + [1] * (n_classes - 1))\n",
    "    weights = weights / np.sum(weights)\n",
    "    labels = np.random.choice(range(n_classes), p=weights, size=size)\n",
    "\n",
    "    # sample source and destination port numbers\n",
    "    spts = np.random.randint(0, 65_535, size=size)\n",
    "    dpts = np.random.randint(0, 65_535, size=size)\n",
    "\n",
    "    # add a few block-like artifacts\n",
    "    min_artifact_size = size / 200\n",
    "    for _ in range(1_000):\n",
    "        tstart = np.random.randint(size)\n",
    "        tlength = np.random.randint(0, min(min_artifact_size, size - tstart - 1))\n",
    "        spts[tstart:tstart + tlength] = np.random.choice(spts)\n",
    "        labels[tstart:tstart + tlength] = np.random.choice(labels)\n",
    "\n",
    "    return pd.DataFrame({\n",
    "        \"time\": timestamps,\n",
    "        \"spt\": spts,\n",
    "        \"dpt\": dpts,\n",
    "        \"label\": labels\n",
    "    })"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "nearby-elder",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = get_synthetic_data(n_classes=5)\n",
    "\n",
    "colors = {0: \"#37474F\", 1: \"#E040FB\", 2: \"#18FFFF\", 3: \"#FFFF00\", 4: \"#C6FF00\"}\n",
    "groups = [\"spt\", \"dpt\"]\n",
    "\n",
    "# to see the interactive plots, install\n",
    "# jupyter labextension install @pyviz/jupyterlab_pyviz\n",
    "\n",
    "shade_network_flow(df, groups, colors, legend=True)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}

requirements.txt

ipython
bokeh>=2
numpy>=1.19
pandas>=1.2
holoviews>=1.14
datashader>=0.12