For more details take a look to https://gmoncarz.github.io/machine_learning_tour/

Machine Learning Tour by Examples by Gabriel Moncarz

This project is a tour for several machine learning procedures showed with examples, applied to one single and specific problem. The main idea is to show, in a simplified way, all the processes evolved when a machine learning model is developed. It shows how the data is gotten from a public data source, preprocessed, transformed and then given to some machine learning models, which produce outputs that later are used to take decisions. Everything is showed step by step, with practical examples implemented in Python 3 throught Jupyter Notebooks.

The problem to model: Earn money in the stock markets

For the entire project one single problem is given, and it will be modeled with different machine learning procedures. It was chosen a realistic project to make it credible and challenging: based on the daily historical quotes of a stock market time serie, use machine learning to decide when to trade it or not. Specifically, the idea is to trade the SPY, which is an ETF that replicates the performance of the S&P 500. Anybody with a stock market account, is able to buy a share of SPY, and it will have a return equivalent to the S&P 500 performance during the investment period. The SPY is a tradable asset and the quotes could be downloaded on Yahoo Finance

The Data

For this project, it is used 19 years of daily SPY quotes: from years 2000 to 2018. The raw data consits of six daily fields:

Date
Open
High
Low
Close
Adjusted close
Volume

The data can be downloaded clicking on the Download Data link of this Yahoo Finance site.

About the project

What this project is about:

Shows the workflow cycle when it is developed machine learning models professionally.
Shows practical implementation of machine learning models in Python, easy to be migrated to other problems.
Shows how is the research process to have good output results.
How new data is generated based on the original data, which helps to have better models. This project shows how from 5 open, low, high, close and volume quotes, it is generated more than 50 derived variables, which could produce better models than only using the 5 underlying quotes.
How to select what are the best variables to use, to have not only a good model, also a simple one.

What this project is not for:

It is not intended to really beat the market. The stock markets has several complications that are out of the scope of this project. For example, there are transaction costs, cost of being short in an asset, etc. The stock market is an excuse for a realistic project, but the focus is on machine learning, but not on the stock markets.
If the reader wants to apply machine learning to stock markets, this is just a starting point.
It is not explained how each machine learning procedure works internally, because there are good courses for each one. This project implements them and the user can see the input, the training process, the output and how that output is used to take decisions.

In a few words, this project is focus on Machine Learning and not in earning money.

Project design

There is one charapter for each machine learning model implemented. Each charapter implements some Jupyter Notebooks which shows the implementation code and the output returned. This implies that all users can download the code and run all the models locally on his own computer. The user can even modify the code to trade other assets, change machine learning parameters, and also implement new trading strategies, like one which trade multiple assets, forex or cryptocurrencies, or even build and ensemble than combines several machine learning models, to leverage the prediction capabilities. I encourage to all users to download this project and based on the developed code, implements his own strategies.

About the implementation details

The whole project is implemented in Python 3. This project calls to the most well-known machine learning libraries which in fact implements each model. The main libraries used are:

Project Source Code

All the project code is located on GitHub