session-manager

Lightweight timestamped session-directory manager for data-processing scripts.

Each time your script runs it gets its own uniquely named folder — no overwritten results, no manual date-stamping. Works on Windows and Linux (pathlib.Path throughout).

Install

pip install session-manager

# always latest from GitHub
pip install git+https://github.com/lukaszplk/session-manager.git

Use cases

1. Save outputs from a data-processing script

Every run produces a fresh folder. Re-run as many times as you like — nothing gets overwritten.

from session_manager import SessionManager

sm = SessionManager("results", name="rna_seq")
# creates: results/rna_seq_2026-05-27_20-00-00/

df.to_csv(sm.file("counts.csv"))
fig.savefig(sm.file("volcano.png"))

plots = sm.subdir("plots")        # creates subdir, returns Path
fig2.savefig(plots / "pca.png")
fig3.savefig(sm / "overview.png") # shorthand

2. Zero-config — bare constructor

No path to decide? Let the library choose:

sm = SessionManager()
# creates: ./sm-sessions/session_2026-05-27_20-00-00/

3. Pipeline chaining — script B always picks up script A's latest output

# script_a.py
sm = SessionManager("results", name="preprocess")
sm.save_params({"lr": 0.01, "epochs": 50})  # writes params.json
df_clean.to_csv(sm.file("clean.csv"))

# script_b.py
sm = SessionManager("results", name="preprocess", create=False)
latest = sm.latest()
df = pd.read_csv(latest / "clean.csv")

4. Browse all past runs

sm = SessionManager("results", name="run", create=False)
for session in sm.list_sessions():           # oldest → newest
    print(session.name)

5. Auto-archive old sessions

Keep the last N sessions active; older ones are moved to an archive folder automatically on each new run:

sm = SessionManager("results", name="run", max_sessions=5)
# once you have 5 sessions, the oldest is moved to results/archive/
# custom archive location:
sm = SessionManager("results", name="run", max_sessions=5, archive_dir="old_runs")

6. Stable symlink for downstream scripts

sm = SessionManager("results", name="run")
sm.symlink_latest()
# creates/updates results/latest → current session
# downstream scripts always read from results/latest/ without calling latest()

Note: On Windows, symlinks require Developer Mode or elevated privileges.

7. Save parameters and environment

sm = SessionManager("results", name="run")
sm.save_params({"lr": 0.01, "dropout": 0.3, "model": "resnet50"})
# → params.json (pretty-printed, non-serialisable values fall back to str)

sm.save_env()
# → environment.txt (pip freeze output for full reproducibility)

# custom filenames
sm.save_params(cfg, filename="config.json")
sm.save_env(filename="requirements.txt")

8. Logger injection — all output in one place

import logging
from session_manager import SessionManager

logger = logging.getLogger("my_pipeline")
sm = SessionManager("results", name="rna_seq", logger=logger)
logger.info("Processing %d samples", len(df))
# library events and your logs share the same handlers/formatters

9. Scratch / temp work

sm = SessionManager.in_temp(name="scratch")
# → /tmp/scratch_2026-05-27_20-00-00/   (Linux/macOS)
# → %TEMP%\scratch_2026-05-27_20-00-00\ (Windows)

10. Experiment tracking

for lr in [0.001, 0.01, 0.1]:
    sm = SessionManager("experiments", name=f"lr_{lr}")
    sm.save_params({"lr": lr})
    model.fit(X_train, y_train, lr=lr)
    json.dump(evaluate(model, X_test), open(sm.file("metrics.json"), "w"))

Options

sm = SessionManager(
    "results",            # omit to use ./sm-sessions/ in cwd
    name="run",
    separator="--",                    # default "_"
    timestamp_format="%Y%m%dT%H%M%S", # default "%Y-%m-%d_%H-%M-%S"
    create=False,                      # default True
    max_sessions=5,                    # default None (no archiving)
    archive_dir="old_runs",            # default base_dir/archive/
    logger=logger,                     # default None (silent)
)

API

Constructor

Methods & properties

Development

git clone https://github.com/lukaszplk/session-manager.git
cd session-manager
pip install -e ".[dev]"
pytest

Work on a feature branch, open a PR against master — CI runs automatically on the PR.