A python demo code for classification and regression by Neural Network (multi-layer perceptron) using Chainer by Shizuo KAJI
MIT Licence
- python 3.6
Install Anaconda if you do not have python 3 on your system.
- Chainer 2.0:
pip install chainer - ChainerUI:
pip install chainerui
Let us try with the famous Iris Data Set. Download iris.data from the web page.
As MLP.py only accepts numerical data, we have to change label names
to integer values.
python csv2mlp.py iris.data.txt -l 4 > iris.csv
does the job. It converts the downloaded file iris.data.txt into a csv compatible with MLP.py
python MLP.py -h
gives a brief description of command line arguments
python MLP.py iris.csv -l 0 -sr 0 -nu 80 80 80 3 -t 5 -a relu -bn
reads iris.csv and use every 5 entries (-t 5) for validation and the rest for training.
Then, it predicts the value of the 0th column (-l 0, 3 classes)
from other columns using a neural network with 80 80 80 3 units (-nu 80 80 80 3)
and with ReLU as an activation function )-a relu) using Batchnormalization (-bn).
Notice that the number of units in the last layer agrees with the number of the classes.
The result is recorded in the file result/result.csv.
python MLP.py iris.csv -l 1 -sr 0 -sc 0 -dr 0 -nu 80 80 1 -t 6 --regression
reads iris.csv and use every 6 entries for validation and the rest for training.
Then, it predicts the value of the 1st column (sepal length)
from columns except for 0th (-sc 0) and 1st.
It plots a graph with matplotlib.
The column indices excluded from training and prediction are specified with -sc
(e.g., -sc 1 3 4 8).
python MLP.py iris.csv -l 1 2 -sr 0 -dr 0.3 -nu 80 80 2 -t 6 --regression
does multi-column (-l 1 2) regression at a time.
Notice that the number of target columns should match the number of units in the last layer (in this case 2).
CSV files should not contain non-numerical values. Remove non-numerical entries (except for the header) using csv2mlp.py (see the above example).