khmer-keyboard 0.1.2

Khmer Keyboard Prediction Engine

Khmer Keyboard Prediction Engine

Khmer keyboard prediction engine for offline/mobile usage.

It supports:

• Next-word prediction (trigram with bigram/unigram fallback)
• Prefix suggestion
• Typo correction
• Smart auto mode (smart) that decides how to respond from user input

Install

From PyPI:

pip install khmer-keyboard

From local source:

pip install -U .

Package Structure

khmer_keyboard/
  train/
    build_lm.py
    build_chargram.py
    export_model.py
  runtime/
    language_model.py
    chargram_embedder.py
    scorer.py
    personalization.py
    engine.py
  model.msgpack

Quick Start

from khmer_keyboard import KhmerKeyboardEngine

# Uses bundled khmer_keyboard/model.msgpack
engine = KhmerKeyboardEngine()

Core APIs

# Prefix suggestion
engine.suggest_prefix("ស", top_n=5)
# ["សូម", "សម្រាប់", "សួស្តី", "សកម្ម", "សាលា"]

# Typo correction
engine.correct_word("កមពជ", top_n=5)
# ["កម្ពុជ", "កម្ពុជា", "កុម្ពុជ", "កម្ពជ", "កពជ"]

# Next-word prediction from context
engine.predict_next("ខ្ញុំ", "ចង់", top_n=5)
# ["ទៅ", "ធ្វើ", "រៀន", "បាន", "សួរ"]

Smart API (Recommended)

smart(...) is the main user-facing API.
It auto-selects behavior:

• If input ends with space: next-word prediction
• If input is partial token: prefix suggestion
• If token looks wrong: typo correction
• If no-space merged phrase is detected: attempts phrase split, then next-word prediction
• If no strong match: safe fallback (never empty list)

Examples:

engine.smart("ខ្ញុំ ចង់ ")   # next-word mode
engine.smart("ស")            # prefix mode
engine.smart("សរឡញ")         # correction/suggestion mode
engine.smart("ខញសរឡញ")       # merged phrase recovery mode

Example Output (Format)

Output is always list[str]:

['ទៅ', 'ធ្វើ', 'ញុំា', 'បាន', 'រៀន']
['សួស្តី', 'សម្រាប់', 'សូម', 'សកម្ម', 'សាលា']

Exact words depend on your trained model and corpus quality.

Use External Model

engine = KhmerKeyboardEngine(model_path="path/to/model.msgpack")