pyhacrf-datamade

0.2.8 verified Fri May 01 auth: no python

Hidden alignment conditional random field (HACRF) for discriminative string edit distance. Current version 0.2.8, infrequent releases.

pip install pyhacrf-datamade

Common errors

error ValueError: Buffer dtype mismatch, expected 'int64' but got 'int32' ↓

cause Incompatible numpy dtype on Windows (int32 default) vs Cython expecting int64.

fix

Explicitly cast arrays: np.array(arr, dtype=np.int64)

error AttributeError: 'HACRF' object has no attribute 'scores_' ↓

cause The 'scores_' attribute was renamed to 'decision_function' in version 0.2.6.

fix

Use model.decision_function(X) instead of model.scores_

Warnings

gotcha Input strings must be passed as lists of characters (or tokens), not as raw strings. For example, pass ['a','b','c'] not 'abc'. ↓

fix Convert strings to list: list('abc')

deprecated The version 0.2.6 changed the API: previously used 'model.fit(X, y, ...)' now requires explicit 'model.fit(X, y)'. Some older examples show deprecated arguments like 'verbose'. ↓

fix Remove deprecated arguments; update to current fit signature.

gotcha The HACRF model expects X to be a list of string pairs? Actually, the input X is a list of strings (each string to align). For pairwise alignment, you need to construct paired examples yourself. ↓

fix For pairwise edit distance, pass two lists: X = list(zip(strings1, strings2)), then model.fit(X, y). But ensure each element is a list of characters.

breaking When upgrading from 0.2.4 to 0.2.6, the internal Cython alignment structure changed; models saved with older versions may not load. ↓

fix Retrain models after upgrading.

Imports

HACRF
wrong
```
from pyhacrf.hacrf import HACRF
```
correct
```
from pyhacrf import HACRF
```
Incorrect submodule path; library uses a simple top-level import.

Quickstart

Minimal example: train HACRF on string pairs and predict.

from pyhacrf import HACRF
import numpy as np

# Example strings as lists of characters
X = [list('abc'), list('abd'), list('abce')]
y = [0, 0, 1]

model = HACRF()
model.fit(X, y)
predictions = model.predict([list('abe')])
print(predictions)