Links:
- Course Github: github.com/DataTalksClub/stoc...
- Registration Form: docs.google.com/forms/d/e/1FA...
- PythonInvest Website: pythoninvest.com/
- (support PythonInvest): buymeacoffee.com/pythoninvest
Timestamps:
[01:34] Module 3 Introduction
[02:08] Agenda for the session
[04:57] System Design View: Module 3 "The Model"
[06:42] Discussing questions from Slido and selected comments from Home Assignment 3 submissions
[18:57] Final Preparations for the Unified Dataset
[31:32] Categorical variables and "dummies" generation in Pandas
[35:31] Correlation Analysis between 200+ features and future positive growth (1 week, boolean)
[40:50] Temporal data split on Train, Validation, and Test sets
[45:35] Similarity of distributions for Train, Validation, Tests sets for a continuous (growth_future_5d) and discrete (is_positive_growth_5d) variables.
[52:20] Introducing Financial Time Series Modeling
[53:13] Manual rules as predictions ("naive" approach)
[59:53] Manual predictions quality on Test dataset
[1:13:33] Autoregressive Integrated Moving Average model (ARIMA) as an example of a statistical prediction for time series
[1:18:40] Binary model: Decision Tree Classifier from the scikit-learn library
[1:27:07] Explainability: obtaining the most important features from trained Decision Tree Classifiers of depth 10 and 20
[1:33:02] Inference on a Decision Tree Classifier and
[1:33:39] Recap
Module 3: Analytical Modeling for Time-Series data
- Framing Hypotheses and Unraveling Time-Series Predictions
- Heuristics and hand rules for practical predictions.
- Predicting time-series data: trends, seasonality, and remainder decomposition.
- Regression techniques for understanding data relationships.
- Binary classification to determine growth direction.
- [Optional] Example of neural networks in analytical modelling.