rahulrajpl / dt-classifier-best-explored

Program trains a decision tree classifier available from sklearn library. Training is undertaken using 3 folded RandomizedSearch cross validation untill specified accuracy is achieved.

Instructions

Following are the two methods to reproduce the results claimed in the report. Preferred method is Method 1.

Method 1 (Easy way)

Pre-requisites on host machine

---

Linux OS (preferably an Ubuntu 18.04) installed with docker container. Same can be installed using instruction at step 1 here. Once the installation is finished, run following command to download the docker image.

docker pull rahulrajpl/clf

This docker image already has the source files. Once the image is downloaded, run the test commands shown in the examples section directly to test the model and evaluate.

Method 2

Pre-requisites on host machine

---

Ensure that Python3.6, python3-tk, virtualenv and pip. If not available, install it using following commands

>> sudo apt-get install python3.6
>> sudo apt-get install python3-tk
>> sudo apt-get install virtualenv
>> sudo apt-get install -y python3-pip

Setting up the source directory

---

1. Extract the zip file submitted

2. Go to the source folder and open a terminal window

3. Activate a virtual environment. This step is optional, however recommended.

   >> virtualenv venv
   >> source venv/bin/activate
   

4. Install all the dependencies using command

   >> pip install -r requirements.txt
   

5. Once the installation is completed, run following command to initialize the training phase. This is an optional step, as a trained model, named 'best.model', is already in the source folder.

   >> python3 train.py
   

6. For testing run following command

   >> python3 test.py <n_samples> <n_iterations> [OPTIONAL model_file]
   

Example usage

---

If training is carried out freshly, then testing can be done as shown below

>> python3 test.py 2000 10

>> python3 test.py 2000 10 best.model 

Above command will test the model for 10 different samples of size 2000 rows. Other examples are shown below

>> python3 test.py 1000 20 best.model
>> python3 test.py 100 10 best.model
>> python3 test.py 1500 20 best.model

Note:

Documentation of source code for training program can be viewed using following command from the terminal

>> pydoc train