sentiment_deepseek_deepinfra.py

import os
import time
import numpy as np
import pandas as pd
from openai import OpenAI

# Set your OpenAI API key and base URL
openai = OpenAI(
    api_key= "mykey",  #Replace with your actual DeepInfra api key
    base_url="https://api.deepinfra.com/v1/openai",
)

stream = False  # Set to True if you want to stream the response
model_used = 'sentiment_deepseek'  # Define the model_used variable

def get_sentiment(symbol, *texts):
    texts = [text for text in texts if text != 0]
    num_text = len(texts)
    text_content = " ".join([f"### News to Stock Symbol -- {symbol}: {text}" for text in texts])

    conversation = [
        {"role": "system",
         "content": f"Forget all your previous instructions. You are a financial expert with stock recommendation experience. Based on a specific stock, score for range from 1 to 5, where 1 is negative, 2 is somewhat negative, 3 is neutral, 4 is somewhat positive, 5 is positive. {num_text} summarized news will be passed in each time, you will give score in format as shown below in the response from assistant."},
        {"role": "user",
         "content": f"News to Stock Symbol -- AAPL: Apple (AAPL) increase 22% ### News to Stock Symbol -- AAPL: Apple (AAPL) price decreased 30% ### News to Stock Symbol -- MSFT: Microsoft (MSTF) price has no change"},
        {"role": "assistant", "content": "5, 1, 3"},
        {"role": "user",
         "content": f"News to Stock Symbol -- AAPL: Apple (AAPL) announced iPhone 15 ### News to Stock Symbol -- AAPL: Apple (AAPL) will release VisonPro on Feb 2, 2024"},
        {"role": "assistant", "content": "4, 4"},
        {"role": "user", "content": text_content},
    ]

    try:
        chat_completion = openai.chat.completions.create(
    #        model='deepseek-ai/DeepSeek-R1-Distill-Llama-70B',
            model='deepseek-ai/DeepSeek-V3',
            messages=conversation,
            temperature=0,
            max_tokens=50,
            stream=stream,
        )

        if stream:
            content = ""
            for event in chat_completion:
                if event.choices[0].finish_reason:
                    print(event.choices[0].finish_reason,
                          event.usage['prompt_tokens'],
                          event.usage['completion_tokens'])
                else:
                    content += event.choices[0].delta.content
            print(content)
        else:
            content = chat_completion.choices[0].message.content
            print(content)
            print(chat_completion.usage.prompt_tokens, chat_completion.usage.completion_tokens)

    except Exception as e:
        print(f"Error: {e}")
        return [np.nan] * num_text

    sentiments = []
    for sentiment in content.split(','):
        try:
            sentiment_value = int(sentiment.strip())
        except ValueError:
            print("content error")
            print("sentiment was: " + str(sentiment.strip()))
            sentiment_value = np.nan
        sentiments.append(sentiment_value)
    return sentiments

def process_csv(input_csv_path, output_csv_path, batch_size=5, chunk_size=1000):
    start_time = time.time()

    # Check if the output file exists and load the last processed row
    if os.path.exists(output_csv_path):
        output_df = pd.read_csv(output_csv_path, 
        on_bad_lines='warn',
        engine='python'
)
        last_processed_row = len(output_df)
    else:
        last_processed_row = 0

    # Read the CSV file in chunks
    chunks = pd.read_csv(input_csv_path, encoding="utf-8", chunksize=chunk_size,
    on_bad_lines='warn', 
    engine='python'     # Print a warning for each skipped line
    )

    for chunk_number, chunk in enumerate(chunks):
        # Skip already processed chunks
        if chunk_number * chunk_size < last_processed_row:
            continue

        chunk.columns = chunk.columns.str.capitalize()
        if model_used not in chunk.columns:
            chunk[model_used] = np.nan

        for i in range(0, len(chunk), batch_size):
            batch = chunk.iloc[i:i + batch_size]
            texts = batch['Lsa_summary'].tolist()
            symbol = batch.iloc[0]['Stock_symbol']  # Extract the stock symbol for the current batch
            sentiments = get_sentiment(symbol, *texts)

            for j, sentiment in enumerate(sentiments):
                if i + j < len(chunk):
                    chunk.loc[chunk.index[i + j], model_used] = sentiment

        # Append the processed chunk to the output file
        chunk.to_csv(output_csv_path, mode='a', header=not os.path.exists(output_csv_path), index=False)

    print(f"Process completed in {time.time() - start_time:.2f} seconds.")

if __name__ == "__main__":
    input_file = 'nasdaq_news_full.csv'  # Replace with your input file path
    output_file = 'sentiment_deepseek_' + input_file
    process_csv(input_file, output_file, batch_size=5, chunk_size=100000)