mirror of
https://github.com/ltcptgeneral/IdealRMT-DecisionTrees.git
synced 2025-09-04 06:17:24 +00:00
Add support for combined datasets and analysis
This commit is contained in:
2
.gitattributes
vendored
Normal file
2
.gitattributes
vendored
Normal file
@@ -0,0 +1,2 @@
|
||||
# force LF for any shell script
|
||||
*.sh text eol=lf
|
4
.gitignore
vendored
4
.gitignore
vendored
@@ -1,3 +1,5 @@
|
||||
data.*
|
||||
__pycache__
|
||||
*.json
|
||||
*.json
|
||||
|
||||
data/*
|
10
README.md
10
README.md
@@ -2,17 +2,21 @@
|
||||
|
||||
Run `pip install -r requirements.txt`
|
||||
|
||||
Run `setup.sh`
|
||||
|
||||
# Tree Generation
|
||||
|
||||
## Download Dataset
|
||||
|
||||
Download the *September 22 2016* dataset from: https://iotanalytics.unsw.edu.au/iottraces.html#bib18tmc
|
||||
Download the *September 22 2016* dataset (or others) from: https://iotanalytics.unsw.edu.au/iottraces.html#bib18tmc
|
||||
|
||||
Rename the file as data.pcap
|
||||
Place these into the `data/tar` folder.
|
||||
|
||||
Run `extract_tars.sh` which will extract and place the `.pcap` files at the corresponding location inside `data/pcap`.
|
||||
|
||||
## Preprocessing Dataset
|
||||
|
||||
Run `ExtractDataset.ipynb`, this will take a few minutes
|
||||
Run `extract_all_datasets.py` which will extract the data from each file in `data/pcap` and turn it into the corresponding `.csv` file inside `data/processed`. This will take a few minutes per file. Combine the data under `data/csv` using `combine_csv.py`. This will overwrite `data/combined/data.csv` which you can use for the decision tree.
|
||||
|
||||
## Training
|
||||
|
||||
|
74
combine.py
Normal file
74
combine.py
Normal file
@@ -0,0 +1,74 @@
|
||||
#!/usr/bin/env python3
|
||||
"""combined.py
|
||||
|
||||
Concatenate every CSV that matches the pattern
|
||||
data/processed/<name>/<name>.csv
|
||||
into a single file:
|
||||
data/combined/data.csv
|
||||
|
||||
The script streams each source CSV in 1‑Mio‑row chunks so memory stays low.
|
||||
Typos in the historic column names (protocl/classfication) are fixed on‑the‑fly.
|
||||
|
||||
Usage
|
||||
-----
|
||||
python combined.py
|
||||
|
||||
You can optionally supply a different root directory:
|
||||
python combined.py --root other/processed_dir --out other/combined/data.csv
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
from pathlib import Path
|
||||
import os
|
||||
import pandas as pd
|
||||
|
||||
CHUNK = 1_000_000 # rows per read_csv chunk
|
||||
|
||||
|
||||
def fix_cols(df: pd.DataFrame) -> pd.DataFrame:
|
||||
"""Rename legacy columns to canonical names."""
|
||||
return df.rename(
|
||||
columns={"protocl": "protocol", "classfication": "classification"}
|
||||
)
|
||||
|
||||
|
||||
def find_source_csvs(proc_root: Path):
|
||||
"""Yield CSV paths that exactly match processed/<name>/<name>.csv."""
|
||||
for sub in sorted(proc_root.iterdir()):
|
||||
if not sub.is_dir():
|
||||
continue
|
||||
target = sub / f"{sub.name}.csv"
|
||||
if target.exists():
|
||||
yield target
|
||||
|
||||
|
||||
def combine(proc_root: Path, out_path: Path):
|
||||
out_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
first_write = True
|
||||
for csv_path in find_source_csvs(proc_root):
|
||||
print(f"→ adding {csv_path.relative_to(proc_root.parent)}")
|
||||
for chunk in pd.read_csv(csv_path, chunksize=CHUNK):
|
||||
chunk = fix_cols(chunk)
|
||||
chunk.to_csv(
|
||||
out_path,
|
||||
mode="w" if first_write else "a",
|
||||
header=first_write,
|
||||
index=False,
|
||||
)
|
||||
first_write = False
|
||||
print(f"✓ combined CSV written to {out_path}")
|
||||
|
||||
|
||||
def main():
|
||||
p = argparse.ArgumentParser(description="Combine processed CSVs into one.")
|
||||
p.add_argument("--root", default="data/processed", help="processed dir root")
|
||||
p.add_argument("--out", default="data/combined/data.csv", help="output CSV")
|
||||
args = p.parse_args()
|
||||
|
||||
combine(Path(args.root).expanduser(), Path(args.out).expanduser())
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
80
extract_all_datasets.py
Normal file
80
extract_all_datasets.py
Normal file
@@ -0,0 +1,80 @@
|
||||
#!/usr/bin/env python3
|
||||
from pathlib import Path
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
from labels import mac_to_label
|
||||
from tqdm import tqdm
|
||||
import os
|
||||
|
||||
ROOT = Path(__file__).resolve().parent
|
||||
PCAP_DIR = ROOT / "data" / "pcap"
|
||||
CSV_DIR = ROOT / "data" / "processed"
|
||||
CSV_DIR.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
BATCH = 100_000 # packets per chunk
|
||||
|
||||
from scapy.all import rdpcap
|
||||
|
||||
|
||||
def process_pcap(pcap_path: str, csv_path: str) -> None:
|
||||
all_packets = rdpcap(pcap_path)
|
||||
|
||||
print("rdpcap done", flush=True)
|
||||
results = []
|
||||
for packet in tqdm(all_packets):
|
||||
size = len(packet)
|
||||
try:
|
||||
proto = packet.proto
|
||||
except AttributeError:
|
||||
proto = 0
|
||||
try:
|
||||
sport = packet.sport
|
||||
dport = packet.dport
|
||||
except AttributeError:
|
||||
sport = 0
|
||||
dport = 0
|
||||
|
||||
proto = int(proto)
|
||||
sport = int(sport)
|
||||
dport = int(dport)
|
||||
|
||||
if "Ether" in packet:
|
||||
eth_dst = packet["Ether"].dst
|
||||
if eth_dst in mac_to_label:
|
||||
classification = mac_to_label[eth_dst]
|
||||
else:
|
||||
classification = "other"
|
||||
else:
|
||||
classification = "other"
|
||||
|
||||
metric = [proto,sport,dport,classification]
|
||||
results.append(metric)
|
||||
results = (np.array(results)).T
|
||||
|
||||
# store the features in the dataframe
|
||||
dataframe = pd.DataFrame({'protocl':results[0],'src':results[1],'dst':results[2],'classfication':results[3]})
|
||||
columns = ['protocl','src','dst','classfication']
|
||||
|
||||
# save the dataframe to the csv file, if not exsit, create one.
|
||||
if os.path.exists(csv_path):
|
||||
dataframe.to_csv(csv_path,index=False,sep=',',mode='a',columns = columns, header=False)
|
||||
else:
|
||||
dataframe.to_csv(csv_path,index=False,sep=',',columns = columns)
|
||||
|
||||
print("Done")
|
||||
|
||||
|
||||
|
||||
def main() -> None:
|
||||
for pcap in sorted(PCAP_DIR.rglob("*.pcap")):
|
||||
rel_csv = pcap.relative_to(PCAP_DIR).with_suffix(".csv")
|
||||
csv_path = CSV_DIR / rel_csv
|
||||
if csv_path.exists():
|
||||
print(f"Skip {rel_csv} (CSV exists)")
|
||||
continue
|
||||
print(f"Processing {rel_csv}")
|
||||
csv_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
process_pcap(str(pcap), str(csv_path))
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
50
extract_tars.sh
Normal file
50
extract_tars.sh
Normal file
@@ -0,0 +1,50 @@
|
||||
#!/usr/bin/env bash
|
||||
# Usage: extract_all.sh SOURCE_DIR TARGET_DIR
|
||||
# For every .tar, .tar.gz, .tgz, .tar.bz2, .tar.xz in SOURCE_DIR:
|
||||
# 1. Create TARGET_DIR/<name>/
|
||||
# 2. If TARGET_DIR/<name>/<name>.pcap already exists, skip the archive.
|
||||
# 3. Otherwise, extract the archive into its own folder.
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
if [[ $# -ne 2 ]]; then
|
||||
echo "Usage: $0 SOURCE_DIR TARGET_DIR" >&2
|
||||
exit 1
|
||||
fi
|
||||
|
||||
src_dir="$1"
|
||||
dst_dir="$2"
|
||||
mkdir -p "$dst_dir"
|
||||
|
||||
# Strip common extensions to recover the base name
|
||||
strip_ext() {
|
||||
local n="$1"
|
||||
n=${n%.tar.gz}; n=${n%.tgz}; n=${n%.tar.bz2}; n=${n%.tar.xz}; n=${n%.tar}
|
||||
echo "$n"
|
||||
}
|
||||
|
||||
shopt -s nullglob
|
||||
for archive in "$src_dir"/*.tar{,.gz,.bz2,.xz} "$src_dir"/*.tgz; do
|
||||
base=$(basename "$archive")
|
||||
name=$(strip_ext "$base")
|
||||
out_dir="$dst_dir/$name"
|
||||
key_file="$out_dir/$name.pcap"
|
||||
|
||||
if [[ -f "$key_file" ]]; then
|
||||
echo "Skipping $archive — $key_file already present"
|
||||
continue
|
||||
fi
|
||||
|
||||
echo "Extracting $archive into $out_dir"
|
||||
mkdir -p "$out_dir"
|
||||
|
||||
case "$archive" in
|
||||
*.tar) tar -xf "$archive" -C "$out_dir" ;;
|
||||
*.tar.gz|*.tgz) tar -xzf "$archive" -C "$out_dir" ;;
|
||||
*.tar.bz2) tar -xjf "$archive" -C "$out_dir" ;;
|
||||
*.tar.xz) tar -xJf "$archive" -C "$out_dir" ;;
|
||||
*) echo "Unknown type: $archive" ;;
|
||||
esac
|
||||
done
|
||||
|
||||
echo "All archives processed."
|
@@ -3,4 +3,5 @@ numpy
|
||||
pandas
|
||||
scikit-learn
|
||||
pydotplus
|
||||
matplotlib
|
||||
matplotlib
|
||||
scipy
|
44
sanity_check/csvdiff.py
Normal file
44
sanity_check/csvdiff.py
Normal file
@@ -0,0 +1,44 @@
|
||||
#!/usr/bin/env python3
|
||||
"""
|
||||
csvdiff.py file1.csv file2.csv
|
||||
Streams both files; prints the first differing line or
|
||||
‘No differences found’. Uses O(1) memory.
|
||||
"""
|
||||
|
||||
import sys
|
||||
from itertools import zip_longest
|
||||
from pathlib import Path
|
||||
|
||||
def open_checked(p: str):
|
||||
print(p)
|
||||
path = Path(p)
|
||||
try:
|
||||
return path.open("r", newline=""), path
|
||||
except FileNotFoundError:
|
||||
sys.exit(f"Error: {path} not found")
|
||||
|
||||
def human(n: int) -> str:
|
||||
return f"{n:,}"
|
||||
|
||||
def main(a_path: str, b_path: str) -> None:
|
||||
fa, a = open_checked(a_path)
|
||||
fb, b = open_checked(b_path)
|
||||
|
||||
with fa, fb:
|
||||
for idx, (ra, rb) in enumerate(zip_longest(fa, fb), 1):
|
||||
if ra != rb:
|
||||
print(f"Files differ at line {human(idx)}")
|
||||
if ra is None:
|
||||
print(f"{a} ended early")
|
||||
elif rb is None:
|
||||
print(f"{b} ended early")
|
||||
else:
|
||||
print(f"{a}: {ra.rstrip()}")
|
||||
print(f"{b}: {rb.rstrip()}")
|
||||
return
|
||||
print("No differences found")
|
||||
|
||||
if __name__ == "__main__":
|
||||
if len(sys.argv) != 3:
|
||||
sys.exit("Usage: csvdiff.py file1.csv file2.csv")
|
||||
main(sys.argv[1], sys.argv[2])
|
600
sanity_check/data_visualization.ipynb
Normal file
600
sanity_check/data_visualization.ipynb
Normal file
File diff suppressed because one or more lines are too long
206
sanity_check/diversity_metrics.py
Normal file
206
sanity_check/diversity_metrics.py
Normal file
@@ -0,0 +1,206 @@
|
||||
#!/usr/bin/env python3
|
||||
"""diversity_metrics.py (fast version)
|
||||
|
||||
Estimate how much diversity each CSV adds without building a giant in‑memory
|
||||
DataFrame. Designed for IoT packet logs with millions of rows.
|
||||
|
||||
Quick summary printed as a GitHub‑style table (requires *tabulate*; falls back
|
||||
to pandas plain text).
|
||||
|
||||
Usage
|
||||
-----
|
||||
python diversity_metrics.py path/to/processed_dir [-r] [--sample 50000]
|
||||
|
||||
Metrics
|
||||
-------
|
||||
ΔEntropy : change in Shannon entropy of *classification* counts
|
||||
ΔGini : change in Gini impurity of the same counts
|
||||
χ² p : Pearson χ² p‑value old vs new classification counts
|
||||
Jaccard : similarity of unique (src,dst) pairs (0 → new pairs, 1 → no new)
|
||||
KS src p : Kolmogorov–Smirnov p‑value, source‑port dist (uses sampling)
|
||||
KS dst p : Kolmogorov–Smirnov p‑value, dest‑port dist (uses sampling)
|
||||
|
||||
Speed tricks
|
||||
------------
|
||||
* No growing DataFrame; we keep Counters / sets / lists.
|
||||
* Ports for KS are *sampled* (default 50 k) to bound cost.
|
||||
* (src,dst) pairs are hashed to a 32‑bit int to reduce set overhead.
|
||||
* pandas reads via **pyarrow** engine when available.
|
||||
"""
|
||||
|
||||
import argparse
|
||||
from pathlib import Path
|
||||
from collections import Counter
|
||||
from typing import List, Set
|
||||
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
from scipy.stats import chi2_contingency, ks_2samp, entropy
|
||||
|
||||
try:
|
||||
from tabulate import tabulate
|
||||
_USE_TABULATE = True
|
||||
except ImportError:
|
||||
_USE_TABULATE = False
|
||||
|
||||
# -----------------------------------------------------------------------------
|
||||
# Helper metrics
|
||||
# -----------------------------------------------------------------------------
|
||||
|
||||
def shannon(counts: Counter) -> float:
|
||||
total = sum(counts.values())
|
||||
if total == 0:
|
||||
return 0.0
|
||||
p = np.fromiter(counts.values(), dtype=float)
|
||||
p /= total
|
||||
return entropy(p, base=2)
|
||||
|
||||
|
||||
def gini(counts: Counter) -> float:
|
||||
total = sum(counts.values())
|
||||
if total == 0:
|
||||
return 0.0
|
||||
return 1.0 - sum((n / total) ** 2 for n in counts.values())
|
||||
|
||||
|
||||
def jaccard(a: Set[int], b: Set[int]) -> float:
|
||||
if not a and not b:
|
||||
return 1.0
|
||||
return len(a & b) / len(a | b)
|
||||
|
||||
# -----------------------------------------------------------------------------
|
||||
# Core analysis
|
||||
# -----------------------------------------------------------------------------
|
||||
|
||||
def analyse(csv_files: List[Path], sample_size: int):
|
||||
"""Return list of dicts with diversity metrics for each added file."""
|
||||
|
||||
# cumulative state (no big DataFrame!)
|
||||
class_counter: Counter = Counter()
|
||||
pair_hashes: Set[int] = set()
|
||||
src_list: List[int] = []
|
||||
dst_list: List[int] = []
|
||||
|
||||
rows = []
|
||||
|
||||
for csv_path in csv_files:
|
||||
df = pd.read_csv(
|
||||
csv_path,
|
||||
engine="pyarrow" if pd.__version__ >= "2" else "c", # fast parse
|
||||
usecols=["protocl", "src", "dst", "classfication"],
|
||||
dtype={
|
||||
"protocl": "uint16",
|
||||
"protocol": "uint16",
|
||||
"src": "uint16",
|
||||
"dst": "uint16",
|
||||
},
|
||||
)
|
||||
# normalise column names
|
||||
df.rename(columns={"protocl": "protocol", "classfication": "classification"}, inplace=True)
|
||||
|
||||
# snapshot previous state
|
||||
prev_class = class_counter.copy()
|
||||
prev_pairs = pair_hashes.copy()
|
||||
prev_src = np.asarray(src_list, dtype=np.uint16)
|
||||
prev_dst = np.asarray(dst_list, dtype=np.uint16)
|
||||
|
||||
# --- update cumulative structures ------------------------------------
|
||||
class_counter.update(df["classification"].value_counts().to_dict())
|
||||
|
||||
# hash (src,dst) into 32‑bit int to save memory
|
||||
pair_ids = (df["src"].to_numpy(dtype=np.uint32) << np.uint32(16)) | \
|
||||
df["dst"].to_numpy(dtype=np.uint32)
|
||||
|
||||
|
||||
# extend port lists (keep small ints)
|
||||
src_list.extend(df["src"].tolist())
|
||||
dst_list.extend(df["dst"].tolist())
|
||||
|
||||
# --- metrics ----------------------------------------------------------
|
||||
# χ² classification
|
||||
chi_p = np.nan
|
||||
if prev_class:
|
||||
all_classes = list(set(prev_class) | set(df["classification"].unique()))
|
||||
old = [prev_class.get(c, 0) for c in all_classes]
|
||||
new = [df["classification"].value_counts().get(c, 0) for c in all_classes]
|
||||
_, chi_p, _, _ = chi2_contingency([old, new])
|
||||
|
||||
# entropy & gini deltas
|
||||
delta_entropy = shannon(class_counter) - shannon(prev_class)
|
||||
delta_gini = gini(class_counter) - gini(prev_class)
|
||||
|
||||
# Jaccard on pair hashes
|
||||
jc = jaccard(prev_pairs, pair_hashes)
|
||||
|
||||
# KS tests on sampled ports
|
||||
ks_src_p = ks_dst_p = np.nan
|
||||
if prev_src.size:
|
||||
new_src = df["src"].to_numpy(dtype=np.uint16)
|
||||
new_dst = df["dst"].to_numpy(dtype=np.uint16)
|
||||
if prev_src.size > sample_size:
|
||||
prev_src_sample = np.random.choice(prev_src, sample_size, replace=False)
|
||||
else:
|
||||
prev_src_sample = prev_src
|
||||
if new_src.size > sample_size:
|
||||
new_src_sample = np.random.choice(new_src, sample_size, replace=False)
|
||||
else:
|
||||
new_src_sample = new_src
|
||||
if prev_dst.size > sample_size:
|
||||
prev_dst_sample = np.random.choice(prev_dst, sample_size, replace=False)
|
||||
else:
|
||||
prev_dst_sample = prev_dst
|
||||
if new_dst.size > sample_size:
|
||||
new_dst_sample = np.random.choice(new_dst, sample_size, replace=False)
|
||||
else:
|
||||
new_dst_sample = new_dst
|
||||
|
||||
ks_src_p = ks_2samp(prev_src_sample, new_src_sample).pvalue
|
||||
ks_dst_p = ks_2samp(prev_dst_sample, new_dst_sample).pvalue
|
||||
|
||||
rows.append(
|
||||
{
|
||||
"File": csv_path.name,
|
||||
"Rows": len(df),
|
||||
"ΔEntropy": round(delta_entropy, 4),
|
||||
"ΔGini": round(delta_gini, 4),
|
||||
"χ² p": f"{chi_p:.3g}" if not np.isnan(chi_p) else "NA",
|
||||
"Jaccard": round(jc, 3),
|
||||
"KS src p": f"{ks_src_p:.3g}" if not np.isnan(ks_src_p) else "NA",
|
||||
"KS dst p": f"{ks_dst_p:.3g}" if not np.isnan(ks_dst_p) else "NA",
|
||||
}
|
||||
)
|
||||
return rows
|
||||
|
||||
# -----------------------------------------------------------------------------
|
||||
# CLI
|
||||
# -----------------------------------------------------------------------------
|
||||
|
||||
def main():
|
||||
ap = argparse.ArgumentParser(description="Evaluate diversity contribution of each CSV (fast version).")
|
||||
ap.add_argument("csv_dir", help="Directory containing CSV files")
|
||||
ap.add_argument("-r", "--recursive", action="store_true", help="Recursively search csv_dir")
|
||||
ap.add_argument("--sample", type=int, default=50_000, help="Sample size for KS tests (default 50k)")
|
||||
args = ap.parse_args()
|
||||
|
||||
root = Path(args.csv_dir)
|
||||
pattern = "**/*.csv" if args.recursive else "*.csv"
|
||||
csv_files = sorted(root.glob(pattern))
|
||||
if not csv_files:
|
||||
print("No CSV files found.")
|
||||
return
|
||||
|
||||
table_rows = analyse(csv_files, args.sample)
|
||||
|
||||
if _USE_TABULATE:
|
||||
print(tabulate(table_rows, headers="keys", tablefmt="github", floatfmt=".4f"))
|
||||
else:
|
||||
print(pd.DataFrame(table_rows).to_string(index=False))
|
||||
|
||||
print(
|
||||
"\nLegend:\n • p-values (χ², KS) < 0.05 → new file significantly shifts distribution (GOOD)"
|
||||
"\n • Positive ΔEntropy or ΔGini → richer mix; near 0 → little new info"
|
||||
"\n • Jaccard close to 0 → many unseen (src,dst) pairs; close to 1 → redundant."
|
||||
)
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
14
setup.sh
Normal file
14
setup.sh
Normal file
@@ -0,0 +1,14 @@
|
||||
#!/usr/bin/env bash
|
||||
# Creates the directory layout:
|
||||
# data/
|
||||
# tar/
|
||||
# pcap/
|
||||
# processed/
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
root="$(cd -- "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||||
|
||||
mkdir -p "$root"/data/{tar,pcap,processed,combined}
|
||||
|
||||
echo "Directory structure ready under $root/data/"
|
Reference in New Issue
Block a user