Spaces:
Running
Running
from __future__ import annotations | |
import os | |
from pathlib import Path | |
CACHE_DIR = Path(os.getenv("CACHE_DIR", ".cache")) | |
CACHE_DIR.mkdir(exist_ok=True, parents=True) | |
DATA_DIR = Path(os.getenv("DATA_DIR", "data")) | |
DATA_DIR.mkdir(exist_ok=True, parents=True) | |
MODEL_DIR = Path(os.getenv("MODEL_DIR", "models")) | |
MODEL_DIR.mkdir(exist_ok=True, parents=True) | |
TOKENIZER_CACHE_DIR = CACHE_DIR / "tokenizer" | |
TOKENIZER_CACHE_DIR.mkdir(exist_ok=True, parents=True) | |
SENTIMENT140_PATH = DATA_DIR / "sentiment140.csv" | |
SENTIMENT140_URL = "https://www.kaggle.com/datasets/kazanova/sentiment140" | |
AMAZONREVIEWS_PATH = DATA_DIR / "amazonreviews.txt.bz2" | |
AMAZONREVIEWS_URL = "https://www.kaggle.com/datasets/bittlingmayer/amazonreviews" | |
IMDB50K_PATH = DATA_DIR / "imdb50k.csv" | |
IMDB50K_URL = "https://www.kaggle.com/datasets/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews" | |
TEST_DATASET_PATH = DATA_DIR / "test.csv" | |
TEST_DATASET_URL = "https://github.com/Tymec/sentiment-analysis/blob/main/data/test.csv?raw=true" | |
SLANGMAP_PATH = DATA_DIR / "slang.json" | |
SLANGMAP_URL = "https://github.com/Tymec/sentiment-analysis/blob/main/data/slang.json?raw=true" | |