Snorkel

0.10.0 verified Sat May 09 auth: no python

A system for quickly generating training data with weak supervision. Uses labeling functions and probabilistic models to create noisy labels. Current version 0.10.0, supports Python >=3.11. Releases are infrequent; maintenance updates only.

pip install snorkel

Common errors

error ValueError: Label model must be fit before calling predict. ↓

cause Calling predict() without first calling fit() on LabelModel.

fix

Call label_model.fit(...) before predict().

error ModuleNotFoundError: No module named 'snorkel.labeling.model' ↓

cause Installed snorkel version <0.9.4 where the model submodule did not exist.

fix

Upgrade snorkel: pip install --upgrade snorkel. Or import from snorkel.labeling import LabelModel if using older version.

Warnings

breaking Snorkel 0.10.0 dropped support for Python <3.11. If your environment uses Python 3.10 or earlier, you must downgrade to snorkel <=0.9.9 or upgrade Python. ↓

fix Use Python >=3.11, or install snorkel==0.9.9 if locked to older Python.

gotcha LabelModel.fit() requires n_epochs to be explicitly set; default is None and will raise error. Many users forget to set n_epochs. ↓

fix Always specify n_epochs (e.g., 500) when calling fit().

deprecated LabelModel renamed from 'LabelModel' to 'LabelModel' (still same) but old path snorkel.labeling.LabelModel still works but discouraged. ↓

fix Use from snorkel.labeling.model import LabelModel.

Imports

LabelModel
wrong
```
from snorkel.labeling import LabelModel
```
correct
```
from snorkel.labeling.model import LabelModel
```
LabelModel was moved to snorkel.labeling.model in v0.9.4, but still accessible via old path for backward compatibility.

LFAnalysis

wrong

from snorkel.labeling import LabelingFunction

correct

from snorkel.labeling import LFAnalysis

LFAnalysis is in snorkel.labeling, not other submodules.

SnorkelClassifier

wrong

from snorkel import SnorkelClassifier

correct

from snorkel.classification import SnorkelClassifier

Must import from snorkel.classification.

Quickstart

Basic Snorkel flow: define labeling functions, generate label matrix, train LabelModel.

from snorkel.labeling import labeling_function, LFAnalysis
from snorkel.labeling.model import LabelModel
import numpy as np

@labeling_function()
def lf_keyword(x):
    # Return 1 if 'keyword' in text else 0
    return 1 if 'keyword' in x else 0

# Example data
data = ['text with keyword', 'no match']
# Apply LFs (simulate L matrix)
L = np.array([[1, 0], [0, 0])
# Train LabelModel
label_model = LabelModel(cardinality=2, verbose=True)
label_model.fit(L_train=L, n_epochs=500, log_freq=100, seed=123)
predictions = label_model.predict(L=L)