Module 1- Part 1- Demystifying timeseries data and modeling (Basics)

Поділитися
Вставка
  • Опубліковано 16 лис 2024

КОМЕНТАРІ • 6

  • @hackerborabora7212
    @hackerborabora7212 6 місяців тому +1

    Oh god the best is back ❤❤ pls do videos about Markov models mamba etc ... Pls i want to study this from you

    • @pedramjahangiry
      @pedramjahangiry  6 місяців тому +1

      I have some knowledge about Mamba, but I'm definitely not an expert! In practice, we still don't have enough evidence to suggest that newer architectures like transformers or even Mamba offer significant benefits considering their implementation costs. Unless we get stable and robust results from these newer architectures, I'll probably hold off on making videos about them. For example, one day I might discuss TKAN (Temporal Kolmogorov-Arnold Networks), which is showing some early promise in the realm of time series forecasting.

  • @alikoohi8265
    @alikoohi8265 6 місяців тому

    Other than financial markets would you please consider examples of industrial time series data?

    • @pedramjahangiry
      @pedramjahangiry  6 місяців тому

      I will provide some notebook templates, you should be able to apply them to any timeseries data.

  • @evau3376
    @evau3376 6 місяців тому

    I was surprised to see how far the future we are going to forecast, ie short term vs long term forecast is not one of the factors to determine the easiness of time series modeling. and also how about the homogenous of volatility?

    • @pedramjahangiry
      @pedramjahangiry  5 місяців тому

      great points.
      - Yes, the forecasting horizon is important, but the main challenge in time series forecasting is the complexity of the data. We can have models that excel at short-term forecasting but perform poorly for long-term forecasts due to the data's complex nature.
      - Homogeneous volatility simplifies modeling, while heterogeneous volatility requires advanced models like GARCH or LSTM to handle changing variance effectively.