Linear Regression is a 42 project where you have to build a program to estimate the price of a car given its mileage.
The idea is to code a gradient descent algorithm based on a dataset of price/mileage cars.
The cost function to minimize is the MSE (for Mean Squared Errors).
There are two programs in the repo. One ./linear_regression.py trains the algorithm; and the other ./predict.py gives a price prediction given the mileage entered by the user.
The main input is a csv file of the form :
km,price
240000,3650
139800,3800
150500,4400
185530,4450
176000,5250
114800,5350
166800,5800
89000,5990
144500,5999
84000,6200
82029,6390
63060,6390
74000,6600
97500,6800
67000,6800
76025,6900
48235,6900
93000,6990
60949,7490
65674,7555
54000,7990
68500,7990
22899,7990
61789,8290
You need to have python 3 installed on your machine.
./linear_regression.py [-h] [-m {BGD,MBGD,SGD}] [-i [0, 5000]] [-a ALPHA]
[-p] [-auto] [-c [0, 100]] [-g [1, 1000]]
data_file
Positional arguments (= required):
| Argument | Description |
|---|---|
| data_file | csv file containing the training examples which will feed the linear_regression algorithm |
Optional arguments :
| Short flag | Long flag | Description |
|---|---|---|
| -h | --help | Show help message |
| -m {BGD, MBGD, SGD} | --method | Type of gradient descent algorithm. Choices are : BGD (Batch GD), MBGD (Mini-Batch GD) or SGD (Stochastic GD) |
| -i [0-5000] | --iterations | Fix number of iterations. Capped at 5000. |
| -a | --alpha | Fix size of Gradient Descent step. |
| -p | --plot | Draw a real time plot of cost function, gradient descent and linear fit as GD advances. If -c is on, plot an analysis of residuals and prediction interval after the training. You can combine with flag -auto for autoscaling. |
| -auto | --autoscale | When -p is True, autoscale the y_axis of the cost function (allows for better visualization in case of divergence) |
| -c [0-100] | --confidence | Undertake an error analysis and use it to build a confidence interval with the given level of confidence |
| -g [1-1000] | --generator | Add randomly generated data points around regression line. |
The predict.py algorithm is simpler.
You just have to execute ./predict.py. Then the prompt will ask you for a mileage. Once you have entered the input and pressed enter, the program will give you the estimated price of your car.
In case you have not trained the regression parameters, the program will warn you and give a prediction of 0 euro.
The ./linear_regression.py program will create a trained_data.json file with the thetas (i.e. regression parameters) recorded for further use in the ./predict.py program.
The ./predict.py program will print on stdout the estimated price of a car with the mileage provided.
