|
--- |
|
license: mit |
|
language: |
|
- ja |
|
tags: |
|
- PyTorch |
|
- Transformers |
|
pipeline_tag: text-classification |
|
--- |
|
|
|
## Japanese Stock Comment Sentiment Model |
|
|
|
This model is a sentiment analysis tool specifically trained to analyze comments and discussions related to Japanese stocks. It is specialized in determining whether a comment has a bearish or bullish sentiment. |
|
For its training, a large collection of individual stock-related comments was gathered, and these were categorized into two main categories: "bullish" and "bearish." This model can serve as a supportive tool for stock investors and market analysts in gathering information and making prompt decisions. |
|
|
|
## How to use |
|
|
|
### Part 1: Model Initialization |
|
|
|
In this section, we'll be initializing the necessary components required for our prediction: the model and the tokenizer. |
|
|
|
```python |
|
import torch |
|
from transformers import AutoModelForSequenceClassification, AutoTokenizer |
|
|
|
# Load the model and tokenizer |
|
model_path = "c299m/japanese_stock_sentiment" |
|
|
|
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") |
|
tokenizer = AutoTokenizer.from_pretrained(model_path) |
|
model = AutoModelForSequenceClassification.from_pretrained(model_path).to(device) |
|
``` |
|
|
|
### Part 2: Text Prediction |
|
|
|
Once our model and tokenizer are initialized, we can move on to predicting the sentiment of a given text. The sentiment is classified into two categories: "bullish" (positive sentiment) or "bearish" (negative sentiment). |
|
|
|
```python |
|
import numpy as np |
|
import torch.nn.functional as F |
|
|
|
# Text for inference |
|
sample_text = "\ |
|
材料良すぎてストップ安、、助けてクレステック、、、\ |
|
" |
|
|
|
# Tokenize the text |
|
inputs = tokenizer(sample_text, return_tensors="pt") |
|
|
|
# Set the model to evaluation mode |
|
model.eval() |
|
|
|
# Execute the inference |
|
with torch.no_grad(): |
|
outputs = model( |
|
inputs["input_ids"].to(device), |
|
attention_mask=inputs["attention_mask"].to(device), |
|
) |
|
|
|
# Obtain logits and apply softmax function to convert to probabilities |
|
probabilities = F.softmax(outputs.logits, dim=1).cpu().numpy() |
|
|
|
# Get the index of the class with the highest probability |
|
y_preds = np.argmax(probabilities, axis=1) |
|
|
|
# Convert the index to a label |
|
def id2label(x): |
|
return model.config.id2label[x] |
|
|
|
y_dash = [id2label(x) for x in y_preds] |
|
|
|
# Get the probability of the most likely class |
|
top_probs = probabilities[np.arange(len(y_preds)), y_preds] |
|
|
|
print(y_dash, top_probs) |
|
``` |
