Swipe to view full content

WEEK 1 WHAT'S IN THERE TASK 1 TASK 2 TASK 3
Day 1
Hey there, excited to start learning?

Welcome to the second-ever edition of our course on Time Series Analysis and Forecasting


First,  we will learn the theoretical aspects of time series forecasting and some basic statistics which will be required afterwards.

What are p-values?

Hypothesis Testing 1

Hypothesis Testing 2 
Theoretical aspects of Time Series forecasting




Autocorrelation and partial correlation?

Autocorrelation - mathematical aspects


Day 2 Today we will learn about prediction intervals and some python tools needed for handling date and time indexed data.

Handbook for handling date time

Prediction Interval Prediction Interval blog
Day 3 Now let us learn the basics of time series and different models used in time series analysis.

White Noise

Log Returns
TS-Basics ADF Test1

ADF Test2

ADF Test3
Day 4 Before applying time series models we need to know data analysis and smoothing methods to get rid of noise.

Analysis of TS data Smoothing methods (upto 40 mins)

Notebook
Day 5 Now ,let's learn how to handle missing values in the data .

Basics of Imputation Imputation following Data Analysis

Swipe to view full content

WEEK 2 WHAT'S IN THERE TASK 1 TASK 2 TASK 3
Day 1
After learning the basics of time series , now we will learn about Facebook Prophet which is an open-source algorithm , used to generate effective time series models.

Basics of Prophet 1

Basics of Prophet 2
Sample Forecasting using Prophet




Multivariate TS forecasting




Day 2 Let's see how to fit a curve with the help of Prophet. The attached notebook will take you through various aspects of Prophet.

Diving deeper into Prophet

Notebook
Multiple Time series modelling
Day 3 Now we will learn time series modelling method ARMA & ARIMA. Here we combine autoregression and moving average method.

ARMA ARIMA
Day 4 Let's see how to write the code for above models. The attached notebook and the follow-along video corresponding to the same notebook will help in this.

Notebook on ARMA/ARIMA Follow-along Video on ARMA/ARIMA
Day 5 Let's now use our knowledge of time series for modelling financial data. Notebook (refer to video in task 2 for more insights) Follow-along Video 

Swipe to view full content

WEEK 3 WHAT'S IN THERE TASK 1 TASK 2 TASK 3
Day 1
We will be using time series techniques for sales forecasting.  Kaggle notebook Corresponding Video on ARMA/ ARIMA









Day 2 Ever wondered how to model a time series in which variance is varying with time? In such a scenario ARCH/GARCH model is what one needs. Theoretical Aspects of ARCH/GARCH

Modelling ARCH/GARCH models




Day 3 Don't want to code a lot? Let's learn about AutoTS(Automatic Time Series), which is a software tool that automates machine learning (AutoML) to build and deploy time series forecasting models. Reader [only up to Ensemble, Caveats and Advice are optional]

Auto TS






Day 4 Today, we will be learning about the Darts library which is available in python and has some great tools for making time series forecasting easier Darts tutorial

Darts tutorial notebook, along with additional information on AutoTS

Video corresponding to the notebook, (watch from 30 mins to 55 mins)




Day 5 Let's have a look at how neural networks and deep learning approach can help in time series analysis. This notebook will take you through models such as LSTM and RNN to accomplish these tasks [Optional] RNN and LSTM

[Optional] RNN and LSTM
Kaggle Notebook explaining RNNs in TSA

Watch the video corresponding to the notebook if required

Swipe to view full content

WEEK 4 WHAT'S IN THERE TASK 1 TASK 2 TASK 3
Day 1
From where we left off, now let's continue to expand our knowledge of using deep learning model to multivariate time series forecasting Theoretical aspects of Multivariate Time Series forecasting using DL [Refresher]


End to End Multivariate Time Series forecasting using DL

Day 2 Sometimes time series data have a natural hierarchical structure and we use Hierarchical time series analysis to analyse and model such data Theoretical aspects Coding Hierarchical Time Series forecasting Part 1

Coding Hierarchical Time Series forecasting Part 2

Day 3 Let's explore how hybrid models are made for time series forecasting with their applications in real-world ES-RNN

N-BEATS

Day 4 Let's see how our time-series models are performing on unseen data using different validation methods.
Also, now we know how to use ML in time series forecasting. Let's see how not to use ML in time series forecasting (Task 3)
Validation Methods - Video




Validation Methods - notebook

Article
Day 5 While it is important to know how to build models, it is also worth noting how we can use the pre-trained models for our problems through transfer learning

Transfer Learning in Time Series
Optional Video







Swipe to view full content

WEEK 5 WHAT'S IN THERE TASK 1 TASK 2 TASK 3
Day 1 A hands on example for CNN implementation would be helpful for building essence of what actually goes into hybrid models

CNN model  CNN model code







Day 2 The two most famous networks in deep learning are CNN and RNN, and Temporal Convolution Network will help combine the two for our task. Temporal Convolution Network

DeepAR




Day 3 You must have used Chat-GPT or something similar which uses a transformer architecture. Today, we will be looking at a model that uses transformers in time series forecasting. Temporal Fusion Transformer Theory Temporal Fusion Transformer Code




LINK DEADLINE INSTRUCTIONS
Assignment




May 17, 2024
N/A

Swipe to view full content

No. WHAT'S IN THERE TASK 1 TASK 2 TASK 3 TASK 4
Topic 1 Neural Networks

Basic  Gradient Descent


Forward and Backward Propagation


Code from Scratch







Topic 2 Convolutional Neural Network Theory

Code from Scratch




Topic 3 Sequential Modeling Recurrent Neural Network(RNN) & Long Short-Term Memory(LSTM) Transformer Long Short-Term Memory(LSTM) Code from Scratch