# relinear

**Repository Path**: williamzjc/relinear

## Basic Information

- **Project Name**: relinear
- **Description**: linear model with relative errors esp. symmetric percentage error.
- **Primary Language**: Python
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2021-07-06
- **Last Updated**: 2021-11-09

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# relinear

## Introduction
linear model with relative errors esp. symmetric percentage error.

see the paper in word format.

## test
just run `script_re.py`

result of one of the numerical experiments

```
               测试误差      训练误差      训练耗时
SMPE-Linear       0.02459   0.02154     2.241
Linear             0.3139    0.2918  0.003044
Ridge              0.2766    0.2464  0.003174
Bayesian           0.3061    0.2908  0.008273
Huber             0.09096   0.08431   0.08062
Theil-Sen          0.2445    0.2278     1.089
Log-SMPE-Linear  0.009494  0.007802     2.093
Log-Linear       0.009539    0.0078  0.003448
Log-Ridge        0.009779   0.00796  0.003489
Log-Bayesian     0.009543    0.0078  0.008904
Log-Huber        0.009668  0.008407   0.05894
Log-Theil-Sen     0.04218   0.03471     1.143
```

## implementation by keras
Just run `re-keras.py`. Since the code is immature, don't import it in any script currently.