Time Series Forecasting With TimeGPT

Time series forecasting presents unique challenges compared to traditional machine learning tasks. Building effective models often requires intricate feature engineering, including windowing and lag creation, yet performance can remain suboptimal, even with sophisticated techniques like LSTMs and GRUs. This is especially true for volatile domains like stock market prediction.

Enter TimeGPT, a cutting-edge foundational model designed to address these limitations. TimeGPT offers state-of-the-art forecasting capabilities, even generalizing well to unseen datasets.

This tutorial explores TimeGPT's architecture, training methodology, and benchmark results. We'll demonstrate how to leverage the Nixtla API to access TimeGPT for forecasting, anomaly detection, visualization, and model evaluation.

Image by Author | Canva

Getting Started with TimeGPT

TimeGPT is accessed exclusively via the Nixtla API (not open-source). This section guides you through API setup and forecasting Amazon stock data.

1. Obtain an API key from dashboard.nixtla.io (account creation required). While currently free, API call limits apply.

2. Configure environment variables within your coding environment (e.g., DataCamp's DataLab). Add the TIMEGPT_API_KEY variable with your key.

3. Install necessary Python libraries:

<code>%%capture
%pip install nixtla>=0.5.1
%pip install yfinance</code>

4. Initialize the Nixtla client using your API key.
5. Download and prepare Amazon stock data:

<code>import pandas as pd
import yfinance as yf
from nixtla import NixtlaClient
import os

timegpt_api_key = os.environ["TIMEGPT_API_KEY"]

nixtla_client = NixtlaClient(api_key=timegpt_api_key)

ticker = 'AMZN'
amazon_stock_data = yf.download(ticker).reset_index()
amazon_stock_data.head()</code>

The data spans from 1997 to the present.

6. Visualize the stock price data:

<code>nixtla_client.plot(amazon_stock_data, time_col='Date', target_col='Close')</code>

7. Perform forecasting (24-day horizon, business-day frequency):

<code>model = nixtla_client.forecast(
    df=amazon_stock_data,
    model="timegpt-1",
    h=24,
    freq="B",
    time_col="Date",
    target_col="Close",
)
model.tail()</code>

8. Plot actual vs. forecasted data (zoomed-in view):

<code>nixtla_client.plot(
    amazon_stock_data,
    model,
    time_col="Date",
    target_col="Close",
    max_insample_length=60,
)</code>

TimeGPT's prediction accuracy is evident.

(The remainder of the original response detailing the Australian electricity demand example is omitted for brevity, but the structure and key elements could be similarly paraphrased and reorganized following the above pattern.)

In conclusion, TimeGPT offers a powerful and accessible solution for time series forecasting, simplifying the process for businesses of all sizes. Its ease of use through the Nixtla API makes advanced forecasting capabilities readily available without requiring extensive machine learning expertise.

The above is the detailed content of Time Series Forecasting With TimeGPT. For more information, please follow other related articles on the PHP Chinese website!