TimeLLM - Time Series Forecasting Model

Показать описание

#TickerTrends #AlternativeData

Adrien Nav

Рекомендации по теме

Комментарии

Adrien, this approach will not work well. We have done exactly that. The problem is LLM weights are a statistical measure of language usages. Even if you fine tuned or prompting it with translated time series vectors, it only has a small % of weight that are relevant to time series. We chose to train a transformer from scratch, how scratch it is, is the key to practicality.

ipsdon

yeah...they dont explain the dataset construction and testing procedure in depth.
if you sample training/testing data at time t and take the next (input, target) sample from t+1 often not much changes, so a persistent estimator (y(t+1) = y(t)) is quite good, i.e. the task is not as hard as when you would move to t+H sample.
Think about temperature and how little it changes now vs next 10 min. Even if that part is fine i.e. with large variation, there is a serious tendency to overfit if you sample like this.
Hope I am wrong

luster

Look up Hurst. I started getting so excited when you realize the chaotic randomness of a particular time FRAME is determinable or predictable via particular frequencies

MlNECRAFT

I don't know if i'm missing it in the paper but, we don't know if these backbone LLMs used some textual interpretation(time series events textualized) of these benchmark time series datasets during pretraining. If so, the backbone models could theoretically look into future where they shouldn't. No way to tell.

onurozkan

Supposedly, stock prices follow a random walk in the short term. However, randomness is fundamentally unpredictable or it may not be randomness at all. Although a LLM may discover hidden variables, good luck with market timing when dealing with Markov chains.

videowatcher-dhil

Yeah im going to have to go with never on this approach for me personally lol. But good luck to anyone that is able to get it to work for them.

Khari

Do you have working tutorial for same method Adrien ?

yogeshagrawal

Great video! Can you please share the codebase which I can try out on my timeseries dataset?

nasiksami

The python code looks like i just creating a prompt with the sequence, just adding the sequence data was not good at all:
You are a helpful assistant that performs time series predictions.
The user will provide a sequence and you will predict the remaining sequence.

Please continue the following sequence without producing any additional text. Do not say anything like 'the next terms in the sequence are, just return the numbers. Sequence:

DennisRutjes

TimeLLM - Time Series Forecasting Model

TimeLLM - Time Series Forecasting Model

TimeGPT: A Foundation Large Time Series Model

Time Series Forecasting of AQI | Ollama | Gemma | TimeLLM | Explanation

[Paper Review] Time-LLM: Time Series Forecasting by Reprogramming Large Language Models

PyData Heidelberg #11 - TimeSeries Forecasting & LLM Langchain

TSLANet For Time Series Forecasting

No Training with TimeGPT Model for Time Series Forecasting !!

LLM-based time series forecasting

TIME LLM TIME SERIES FORECASTING BY REPROGRAMMING LARGE LANGUAGE MODELS （Monash University & An...

Time Series Talk : Autoregressive Model

TimeGPT: Forecasting Made Easy

Chronos: Learning the Language of Time Series with Abdul Fatir Ansari - 685

LLM Explained | What is LLM

An LLM journey speed run: Going from Hugging Face to Vertex AI

LLM with Time Series 2 [Time series forecasting using LLMs]

[Paper Review] Temporal Data Meets LLM - Explainable Financial Time Series Forecasting

AI Stock Price Prediction Using Large Language Models in Python

Research Paper VISIONTS: VISUAL MASKED AUTOENCODERS ARE FREE-LUNCH ZERO-SHOT TIME SERIES FORECASTERS

TimeGPT: A foundation model for time series 2024-03-19

Why LLM's Can't Trade Stocks

Can Large Language Models Beat the Market? Testing AI Time Series on Crypto Prices in Python

Are Language Models Actually Useful for Time Series ForecastingVirginia & UW 2024

Epoch, Batch, Batch Size, & Iterations

ACL 2024 Keynote: Can LLMs Reason & Plan?