This is the repository of the sorce code for CIS 700 Neural Program Learning's semester project. The purpose of this project is to use the architecture of NTM to solve 4 different tasks. Each team member is correpond to one task alone. The task may came from internet and it may varies from its original version. This project is modified from the original project at here.

• Dr. Garrett Katz(gkatz01@syr.edu)

• Ruipeng LIU(rliu02@syr.edu)
  • Maximum sub array sum problem
  • Branch name: rliu02
• Qinwei HUANG (qhuang18@syr.edu)
  • Merge intervals
  • Branch name: qhuang18
• Chi ZUO (czuo02@syr.edu)
  • Palindrome numbers
  • Branch name: czuo02
• Yuchao XU (yxu219@syr.edu)
  • Valid Parity
  • Branch name: yxu219

In order to run this project, please make sure that the hosting machine having a python3 installed. Being able to build a virtual environment or any other isolated environment is preferred since some of the packages that this project used may not be the most up to date version.

Again it is highly recommended to install an isolated environment. This instruction assume the installation happens in virtual environment.

First, activate virtual environment. Then go to the root folder of the project, where the requirements.txt is located. Execute the following command:

pip install -r requirements.txt

There may be issues regarding different package name and versions accross different OS, in that case, please make sure that the following packages are properlly installed:

• attrs==19.1.0
• matplotlib==3.3.4
• numpy==1.19.5
• torch==1.8.1

First, select the corresponding branch that aims to solve a particular problem. Pull the branch to the local. Make sure that all the dependencies are installed.

The following command are assume that the virtual environment is activated.

To generate training and testing data, please execute the following command at the root folder of the project:

python data_generator.py

If successfully executed, 3 files named train.txt, test.txt, test_var.txt corresponding to trainig data, testing data with the same input size with the trainig data and the testing data with different input size with the trainig data should be presented in the same directory.

To train the model, please make sure that the file train.txt exist in directory CIS700_proj/data/. Go to the root directory of this project where train.py is located, simply execute the following command:

python train.py

If the program execute correctly, a log at the end states Done training should be presented. A file under CIS700_proj/model/ is presented. And finally a file named train_loss.txt is presented in the root directory of the project.

To test the model, please make sure that test.txt, test_var.txt exist in directory CIS700_proj/data/ and a model file exist in directory CIS700_proj/model/. Go to the root directory where test.py, test2.py exist. To run the test with the same input size of the training file, execute:

python test.py

To run the test with different input size of the training file, execute:

python test2.py

If the program execute correctly, a file named test_loss.txt (or test2_loss.txt for test2.py) will be presented in the root folder.

Please make sure that you have a GUI. This instruction may not work for a pure text use interface. To plot the losses, please make sure that all train_loss.txt, test_loss.txt, test2_loss.txt exist in the root directory. Execute command:

python learning_curve.py

If execute correctly, a graph plotting 3 curves corresponding 3 losses should be presented.

Some configurations can be accessed and modified via CIS700_proj/data/config.py. Within this file, it is possible to change the following attributes of the project:

• Upper and lower bound of each element in the training data
• EOS's representation, make sure that this representation cannot be accessed by all possible elements in input and output
• Size of input
• Encoding length of each element both input and output (an unsigned integer's binary representation)
• Threshold for encoding. (Shift the encoding data to non-negative)
• Training, testing and variate testing data size
• A portion of neural network configuration

Other configs of the model needs to be accessed within the code.

There is a sample trained model located within the folder CIS700_proj/model/ that is trained under the current configuration. Corresponding training and testing datasets are located in CIS700_proj/data/. If you would like to train another model, please make sure that the training data is available and remove the model first. If you wish to regenerate the training and testing data, please make sure that the original training and testing datasets are removed.

A set of sample train and test loss data also exist in the root directory, if you wish to generate your own loss data, please remember to remove train_loss.txt, test_loss.txt, test_loss2.txt