kiwipiepy Model

0.23.0 · active · verified Fri Apr 17

kiwipiepy-model provides the language models necessary for kiwipiepy, a fast and accurate Korean morphological analyzer. It is a data-only package, implicitly installed as a dependency when you install kiwipiepy. The current version is 0.23.0, and it generally updates in lockstep with major versions of the core kiwipiepy library to provide the latest model improvements.

Common errors

```
ModuleNotFoundError: No module named 'kiwipiepy'
```
cause You likely installed `kiwipiepy-model` but not `kiwipiepy` itself, or have not installed either.

fix Install the main library: `pip install kiwipiepy`.
```
TypeError: Kiwi() got an unexpected keyword argument 'typo_correct_options'
```
cause In kiwipiepy v0.23.0+, `typo_correct_options` is no longer a constructor argument. It's now passed to analysis methods.

fix Pass `typo_correct_options` to `kiwi.tokenize()`, `kiwi.split_into_sentences()`, etc., instead of `Kiwi()`.
```
TypeError: Kiwi() got an unexpected keyword argument 'oov_score_mode'
```
cause In kiwipiepy v0.23.0+, OOV handling is configured via the `oov_handling` parameter, replacing older arguments like `oov_score_mode`.

fix Remove `oov_score_mode` from the `Kiwi` constructor. Use the new `oov_handling` parameter (e.g., `oov_handling='ngram'`) when initializing `Kiwi()` or in analysis methods.

Warnings

breaking The `typo_correct_options` parameter for typo correction has been moved from the `Kiwi` constructor to the `tokenize`, `split_into_sentences`, etc., methods.
Fix: Instead of `Kiwi(typo_correct_options=...)`, use `kiwi.tokenize(..., typo_correct_options=...)` or `kiwi.split_into_sentences(..., typo_correct_options=...)`.
breaking The way to specify OOV (Out-Of-Vocabulary) handling has changed from boolean flags or simple modes to a more flexible `oov_handling` parameter that accepts specific string values or configurations in the `Kiwi` constructor.
Fix: Remove deprecated OOV-related parameters (e.g., `oov_score_mode`) from `Kiwi()` constructor. Use `oov_handling='length_based'` or `oov_handling='ngram'` in the `Kiwi()` constructor, or pass it to analysis methods for specific calls.
gotcha The `kiwipiepy-model` package only provides the model files. You must install and import from the `kiwipiepy` package to use the morphological analyzer.
Fix: Ensure you `pip install kiwipiepy` and `from kiwipiepy import Kiwi`. Do not attempt to import classes directly from `kiwipiepy_model`.

Install

pip install kiwipiepy Recommended: Installs kiwipiepy and its model
pip install kiwipiepy-model Install model package only (not typically needed)

Imports

Kiwi
wrong:
```
from kiwipiepy_model import Kiwi
```
correct:
```
from kiwipiepy import Kiwi
```
The 'kiwipiepy-model' package provides model data, but the main 'Kiwi' class is imported from 'kiwipiepy'.

Quickstart

Initialize the Kiwi morphological analyzer. The default model is automatically loaded from the kiwipiepy-model package. This example demonstrates basic tokenization and sentence splitting, incorporating the new `oov_handling` parameter.

from kiwipiepy import Kiwi

# Initialize Kiwi with default model (loaded from kiwipiepy-model)
kiwi = Kiwi(oov_handling='ngram') # Example with new oov_handling

text = "안녕하세요, 저는 키위입니다."
tokens = kiwi.tokenize(text)
for token in tokens:
    print(f"Word: {token.form}, Tag: {token.tag}, Start: {token.start}, End: {token.end}")

sentences = kiwi.split_into_sentences("어제는 비가 왔어요. 오늘은 맑을 예정입니다.")
for sentence in sentences:
    print(f"Sentence: {sentence.text}")
    for token in sentence.tokens:
        print(f"  {token.form}/{token.tag}")

view raw JSON →