Description

Small Example of a simple linear regression script.

DISCLAIMER: This script is for demonstrative propouses only, it should not be used in production without the proper analysis of the variables to predict, including but not limited to their autocorrelation/multicollinearity, correlation and or seasonality.

Install

This project requires
. It is recommended to create a virtual enviorment to use it.

git clone --depth 1 https://github.com/hriva/sample-linear-regression.git 
cd sample-linear-regression

# Create virt env
virtualenv -p python3 sample-linear-regression
source sample-linear-regression/bin/activate
cd sample-linear-regression
pip3 install -r requirements.txt

# Run
chmod +x src/bitcoin-ethereum.py
./src/bitcoin-ethereum.py

Explanation

Linear Regression

Small linear regression sample implementation. For this example we are using the Bitcoin Prices in a Monthly basis as the dependant variable. i.e., the variable we want to predict. And we use the Ethereum Prices in a Montly basis as the independant variable. i.e., the variable we are using to predict the Bitcoin price.

DISCLAIMER: this is simplified given that yahoo offers good quality data. Usually you need more steps to clean and wrangle data.

1. Import libraries

import pandas as pd
from sklearn.linear_model import LinearRegression
import matplotlib.pyplot as plt

2. Load the data from csv files.

# Ingest
## Data paths.
DATA = "src/assets/BTC-USD.csv" 
DATA1 = "src/assets/ETH-USD.csv"

## Make pandas read the data.
df1 = pd.read_csv(DATA, index_col="Date", parse_dates=True).sort_index(ascending=True)
df2 = pd.read_csv(DATA1, index_col="Date", parse_dates=True).sort_index(ascending=True)

Ingests the data fetched from Yahoo Finance (The data has no blank values). During the import, the data is formated to a time series by setting the Dates as the index. The index is then sorted as ascend given that for linear regresions these need to be from older to newest.

df1.head()
df2.head()

	Open	High	Low	Close	Adj Close	Volume
Date
2019-02-01	107.147682	165.549622	102.934563	136.746246	136.746246	101430995445
2019-03-01	136.836243	149.613235	125.402702	141.514099	141.514099	138882123600
2019-04-01	141.465485	184.377853	140.737564	162.166031	162.166031	204556824026
2019-05-01	162.186554	287.201630	159.660217	268.113556	268.113556	314349041886
2019-06-01	268.433350	361.398682	229.257431	290.695984	290.695984	270589672710

3. Preprocess the data for the regression.

## Get the correlation
df1["Adj Close"].corr(df2["Adj Close"])

0.9187847614681434

## Get necesary predictive values only.
# Use double brackets to avoid sending pandas.core.series.Series instead of DataFrame
dfx = df2[["Adj Close"]]  # Load the Etherum price as x (Independant)
dfy = df1[["Adj Close"]]  # Load the Bitcoin price as y (Dependant)

Viewing the data shows the varying measures for the exchange prices. We need to drop all of them except the "Adj Close".

Notice that unlike the correlation. We fetch the columns using double brackets. This is to avoid getting errors in scikit-learn given that we are using 1 variable as preditor instead of a multy plexed array (a matrix).

4. Split the sets

train_size = 0.8  # use 80 percent to train the regression
if dfx.shape[0] != dfy.shape[0]:
    print("Sample Sizes ERROR")
    exit
x_train_size = round(dfx.shape[0] * train_size)  # We only need the rows.
x_test_size = x_train_size
y_train_size = round(dfx.shape[0] * train_size)  # We only need the rows.
y_test_size = y_train_size

x_train, x_test = dfx.iloc[:x_train_size], dfx.iloc[x_test_size:]
y_train, y_test = dfy.iloc[:y_train_size], dfy.iloc[y_test_size:]

Why not use all the series for the regression? To avoid overfitting.

5. Regression

# Regression
regressor = LinearRegression()
regressor.fit(X=x_train, y=y_train)

Create a Linear Regression instance and then fit it to the linear regression we need.
x is the Ethereum price.
y is the Bitcoin price.

6. Predict

y_pred = regressor.predict(x_test)
print(y_pred)

[[29754.22137026]
 [30433.23886914]
 [30398.68553649]
 [31129.1065507 ]
 [30176.4599037 ]
 [27572.75954253]
 [27888.41823535]
 [29685.41814795]
 [32605.41128462]
 [35436.57757274]
 [35799.63595354]
 [34648.96926019]]

Create an array with the predictions for the test (Validation) set.
Print the predictions.

7. Plot.

# Training Sets
plt.scatter(x_train, y_train, color="red")
plt.plot(x_train, regressor.predict(x_train), color="blue")
plt.title("Bitcoin vs Ethereum")
plt.xlabel("Ethereum Closing Price")
plt.ylabel("Bitcoin Closing Price")
plt.show()


# Test set
plt.scatter(x_test, y_test, color="red")
plt.plot(x_train, regressor.predict(x_train), color="blue")
plt.title("Bitcoin vs Ethereum")
plt.xlabel("Ethereum Closing Price")
plt.ylabel("Bitcoin Closing Price")
plt.show()