berniwal / deeplearningproject

Folder Structure

Dataset

This folder contains the Datasets used for training and testing

train-easy

This folder contains the Datasets used for training and interpolation testing

train-medium

This folder contains the Datasets used for testing on medium extrapolation tests

train-hard

This folder contains the Datasets used for testing on hard extrapolation tests

Models

This folder contains all the code which is able to run. It contains two notebooks, one for each Subproblem mentioned in the paper.

arithmetic_add_or_sub_images

This folder contains images which were used in the paper to show some examples

best_arithmetic__add_or_sub_model

This folder contains the checkpoint of the (best) model for the arithmetic__add_or_sub problem used to measure the performance of the model.

best_calculus_differentiate_model

This folder contains the checkpoint of the (best) model for the calculus_differentiate problem used to measure the performance of the model.

calculus__differentiate_images

This folder contains images which were used in the paper to show some examples

Writing

This folder contains the proposal and the paper itself.

How to Run the Code

• If you have not downloaded the Repository From Github then you have to download the Dataset folder from https://github.com/berniwal/DeepLearningProject/ and replace it as there was not enough space to compress it to a 100MB zip file. Otherwise if downloaded from Github you can skip this step.
• Go to the Models folder and select the Notebook for which subproblem you want to run the code.
• Make sure you have installed the required libraries in the first import statement (tensorflow 2.x) otherwise you will not be able to run the code.
• If you made sure you have installed those libraries, then you should be able to execute the notebook without any further considerations.

Related Work

Based on the Paper 'Analysing Mathematical Reasoning Abilities of Neural Models'

https://arxiv.org/pdf/1904.01557.pdf

Link to Mathematics Dataset

https://github.com/deepmind/mathematics_dataset

Link to Pre-Generated Dataset

https://console.cloud.google.com/storage/browser/mathematics-dataset?pli=1

Link to used tutorial about Neural Machine Translation with Attention

https://www.tensorflow.org/tutorials/text/nmt_with_attention