Try it in five minutes

This page takes you from nothing to a printed result for two models: picket-forest (CAN bus intrusion detection) and scout-large (jobs, pains and gains in mobility texts). It works on macOS and Linux. Every model needs Python 3.12 or newer, and the steps below give you one in a minute.

The install lines and programs on this page are the ones on the model cards. If you read a card instead, you will find the same.

1. One environment with Python 3.12

Install uv (for example brew install uv on macOS), then make an empty folder with its own environment:

mkdir try-zoo && cd try-zoo
uv venv --python 3.12
source .venv/bin/activate
python --version

The last line must say Python 3.12 or newer. uv downloads Python 3.12 if your machine only has an older one. Without uv, python3.12 -m venv .venv does the same, and you then use pip install where this page says uv pip install.

2. picket-forest

Install it. The edge extra holds what the model needs (numpy, scikit-learn) and no deep-learning framework, roughly 150 to 250 MB:

uv pip install "mobility-model-zoo[edge] @ git+https://github.com/mhabedank/mobility-model-zoo@picket-forest/v0.2.0"

Save this program as demo.py. It downloads the model, an example capture of CAN frames and the calibration table of the example vehicle from the model repository, scores the frames and prints the result:

from huggingface_hub import hf_hub_download

from mobility_model_zoo.security.can_ids import picket_forest as pf

model = pf.load_model(hf_hub_download("mobility-model-zoo/picket-forest", "model.joblib", revision="v0.2.0"))
table = pf.load_table(hf_hub_download("mobility-model-zoo/picket-forest", "calibration_table.json", revision="v0.2.0"))
capture = pf.load_capture(hf_hub_download("mobility-model-zoo/picket-forest", "examples/01-fabrication.json", revision="v0.2.0"))
scores, alarms = pf.score(capture, table, model)
print(pf.summary(capture, scores, alarms))

Run it:

python demo.py

You see something like this:

(the model page shows what the program prints)

The model flags frames that break the rules calibrated on one vehicle, and the alarm stage raises an alarm when several flagged frames follow each other. To use it on your own vehicle, calibrate it on benign traffic of that vehicle first; the model page shows how.

3. scout-large

This is the text model. Its extra holds PyTorch and Transformers, so the install is larger than the first one (several hundred MB). Use the same environment:

uv pip install "mobility-model-zoo[text] @ git+https://github.com/mhabedank/mobility-model-zoo@scout-large/v0.1.2"

Save this program as demo_text.py, then run it with python demo_text.py:

import json

from mobility_model_zoo.productdev.jtbd.span import SpanExtractor

model = SpanExtractor.from_pretrained("mobility-model-zoo/scout-large", revision="v0.1.2")
print(json.dumps(model.extract("Ich wohne am Stadtrand und arbeite im Schichtdienst. Nach 22 Uhr fährt kein Bus mehr, also nehme ich das Auto, obwohl ich lieber Bahn fahren würde. Wenn es einen Nachtbus gäbe, würde ich das Auto verkaufen."), ensure_ascii=False, indent=2))

The first run downloads the model (about 2.2 GB, once) and takes a few seconds to load; with a fast connection plan a few minutes for this step. The result is JSON: for every job, pain or gain in the text, the exact sentence it was found in (quote, with its position), a score and a few attributes. An empty list is a valid result.

When something goes wrong

requires a different Python: 3.11.x not in '>=3.12'. The environment is not Python 3.12. Run python --version inside the activated environment. Create it again with uv venv --python 3.12.

IPython starts with another Python version than the environment. A banner such as Python 3.11.7 after you activated a 3.12 environment means ipython comes from another interpreter, and it cannot see the packages you just installed (No module named 'mobility_model_zoo'). Run your program with python demo.py, or install IPython into the environment (uv pip install ipython) and start it with python -m IPython.

No module named pip. Environments made by uv do not contain pip. Use uv pip install ..., or create the environment with uv venv --seed.

command not found: from (or import). The program was pasted into the shell. Save it as a file (demo.py) and run it with python demo.py.

No module named 'joblib', 'pandas' or 'sklearn' after the install. You installed without the extra on the install line, or into another environment. Copy the install line from the card as it is, and check which python points into .venv.

Windows. Not tested. Nothing in the picket-forest path needs a compiler, so it should work in a Python 3.12 environment there too.

Where to go next

The model pages show quality, speed, limits and the worked examples of every model: the model list. For picket-forest on a microcontroller, the card also shows the C code for the device.