metadata
library_name: transformers
tags: []
AfriSenti Swahili Sentiment Regressor Description
Takes a text and predicts the sentiment value between -1 (Negative) to 1 (Positive) with 0 being Neutral.
Regression Value Description:
| Value | Sentiment |
|---|---|
| -1 | Negative |
| 0 | Neutral |
| 1 | Positive |
How to Get Started with the Model
Use the code below to get started with the model.
import math
import torch
import pandas as pd
from transformers import AutoModelForSequenceClassification, AutoTokenizer
BATCH_SIZE = 32
ds = pd.read_csv('test.csv')
BASE_MODEL = 'HausaNLP/afrisenti-swa-regression'
device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
model = AutoModelForSequenceClassification.from_pretrained(BASE_MODEL)
nb_batches = math.ceil(len(ds)/BATCH_SIZE)
y_preds = []
for i in range(nb_batches):
input_texts = ds[i * BATCH_SIZE: (i+1) * BATCH_SIZE]["tweet"]
encoded = tokenizer(input_texts, truncation=True, padding="max_length", max_length=256, return_tensors="pt").to(device)
y_preds += model(**encoded).logits.reshape(-1).tolist()
df = pd.DataFrame([ds['tweet'], ds['label'], y_preds], ["Text", "Label", "Prediction"]).T
df.to_csv('predictions.csv', index=False)