Udemyexpertfr

Prédiction des séries temporelles en deep learning - Partie2

Learn advanced machine learning with Keras, TensorFlow, and Random Forest. Master hyperparameter optimization for powerful models.

4.8(10)
217students
9h 42m
Certificate
Alexandre Bourrieau
Alexandre BourrieauEnseignant

Course Content

Prédiction des séries temporelles en deep learning - Partie2 is an advanced Udemy course taught by Alexandre Bourrieau. It runs 9h 42m, is taught in French, and includes a certificate of completion. Learners rate it 4.8 out of 5 from 10 reviews, with 217 students enrolled. It is a paid course, first published August 2021.

Advanced Time Series Forecasting with Deep Learning

Expand your expertise beyond the fundamentals of time series prediction with deep learning! This advanced course builds upon the concepts introduced in "Time Series Forecasting with Deep Learning - Part 1," diving into cutting-edge models and techniques for highly accurate predictions. Learn to implement and optimize state-of-the-art architectures using Python and industry-leading libraries.

Key Skills You'll Master:

  • Advanced Deep Learning Models: Explore complex models like DSTP-RNN, HRHN, STAM, and Wavenet, all designed for superior time series forecasting performance.
  • Python Implementation: Gain hands-on experience coding these models using TensorFlow, Keras, Pandas, NumPy, and Scikit-learn.
  • Hyperparameter Optimization: Utilize Raytune to efficiently optimize model hyperparameters, achieving maximum prediction accuracy.
  • Feature Importance & Selection: Apply Random Forest algorithms and VSURF (using R on Google Cloud) to identify and prioritize the most impactful variables in multivariate time series data.
  • Error Compensation with PID Controllers: Implement PID controllers to enhance model performance and improve prediction accuracy.
  • Cloud Computing for Data Analysis: Leverage Google Colab for online Jupyter notebooks and Google Cloud for powerful R-based variable selection using VSURF.

Course Highlights:

  • DSTP-RNN: A powerful multivariate Seq2Seq model that addresses the limitations of DA-RNN by incorporating spatio-temporal attention.
  • HRHN: Leverages Recurrent Highway Networks (RHN) and 1D convolutional layers to overcome convergence issues in deep recurrent neural networks.
  • PID Controller Integration: Learn to incorporate PID controllers into DSTP-RNN and HRHN models for improved performance.
  • Raytune for Hyperparameter Tuning: Employ advanced optimization algorithms like Hyperband and Bayesian optimization to fine-tune model parameters.
  • Random Forest for Variable Selection: Utilize Random Forest to identify the most relevant variables in complex time series datasets, reducing computational burden.
  • VSURF on Google Cloud: Master the use of VSURF in R on Google Cloud for robust variable selection and feature interpretation.
  • STAM: Causal & Evolutionary Modeling: Explore the STAM model, a recent attention-based architecture known for its causal and evolutionary properties.
  • Wavenet for Time Series Prediction: Discover how to adapt the Wavenet architecture, originally designed for audio generation, for improved time series forecasting.

Prerequisites:

This course is a continuation of "Time Series Forecasting with Deep Learning - Part 1." Prior experience with deep learning is recommended. A basic understanding of mathematics and programming is helpful.

Support & Resources:

I'm here to support your learning journey! Don't hesitate to ask questions – I'll respond promptly. Your motivation is key to success.

Course Duration: Over 9 hours of video content.

Get ready to take your time series forecasting skills to the next level!

$ whoami — the devs behind tutorialsearch.io

Want AI in your product? Hire the team behind TutorialSearch.io

LLM features, RAG, agents and automation wired into a real product — built by the devs who shipped this site.