Set Up Kronos

I want to set up Kronos, the open-source AI model that forecasts financial markets, and run my first forecast. I am not a developer, so do everything for me and explain each step in plain English as you go.

Here is the repo: https://github.com/shiyu-coder/Kronos

Please:
1. Check that I have Python 3.10 or newer installed. If I don't, tell me exactly how to install it for my operating system.
2. Clone the Kronos repo into a new folder called "kronos" in my current directory.
3. Create a virtual environment inside it and install everything in requirements.txt, plus yfinance so we can pull real market data.
4. Write a script called forecast.py that:
   - Downloads the last 400 daily candles for a ticker I choose (default: AAPL) using yfinance
   - Loads the Kronos-small model and the Kronos-Tokenizer-base from Hugging Face
   - Feeds in the historical open, high, low, close, and volume
   - Forecasts the next 30 trading days
   - Saves a chart called forecast.png that shows the real price history and the predicted path on the same plot
5. Run forecast.py and open forecast.png so I can see the result.

If anything errors out, read the error, fix it yourself, and try again. Don't ask me to fix code. When you're done, tell me in plain English what the chart is showing, and remind me this is a forecast, not financial advice.

from model import Kronos, KronosTokenizer, KronosPredictor
import pandas as pd
import yfinance as yf

# 1. Load the model (downloads from Hugging Face the first time)
tokenizer = KronosTokenizer.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
model = Kronos.from_pretrained("NeoQuasar/Kronos-small")
predictor = KronosPredictor(model, tokenizer, max_context=512)

# 2. Pull real market data (last 400 daily candles for Apple)
df = yf.download("AAPL", period="400d", interval="1d").reset_index()
df.columns = [c.lower() for c in df.columns]

lookback = 400
pred_len = 30
x_df = df.loc[:lookback - 1, ['open', 'high', 'low', 'close', 'volume']]
x_timestamp = df.loc[:lookback - 1, 'date']
y_timestamp = pd.date_range(x_timestamp.iloc[-1], periods=pred_len + 1, freq='B')[1:]

# 3. Forecast the next 30 days
pred_df = predictor.predict(
    df=x_df,
    x_timestamp=x_timestamp,
    y_timestamp=y_timestamp,
    pred_len=pred_len,
    T=1.0,
    top_p=0.9,
    sample_count=1,
)

print(pred_df)

1. Change the ticker and timeframe
"Edit forecast.py to forecast TSLA on the hourly chart instead of AAPL daily. Pull the last 512 hourly candles and forecast the next 48 hours."

2. Backtest it against reality
"Re-run the forecast, but cut off the data 30 days ago. Then plot the forecast against what actually happened over those 30 days so I can see how close it got."

3. Compare a basket of tickers
"Run the forecast for AAPL, NVDA, MSFT, and AMZN. Make one chart per ticker and tell me which one Kronos expects to move the most over the next 30 days."

4. Read the volatility, not just the price
"Run the forecast with sample_count set to 30 so we get a range of possible paths. Shade the area between the highest and lowest paths so I can see how uncertain the model is."

5. Upgrade to the bigger model
"Swap Kronos-small for Kronos-base and re-run the same forecast. Tell me if the prediction changed and roughly how much slower it was."