Recommend using python3 and a virtual env.

conda create -n myenv python=3.6
source activate myenv
pip install -r requirements.txt

Given kaggle handwritten math symbols dataset prepare data for future analysis.

We will used the handwritten math symbols dataset from kaggle. Download it here.

Here is the raw structure of the data:

extracted_images/
    log/
        filename.jpg
        ...
    sigma/
        filename.jpg
        ...

The dataset contains 375974 images belonging to 82 labels.

Once the download is complete, move the dataset into MathMnist/data:

mv path/to/extracted_images/ path/to/MathMnist/data/

Run the script build_dataset.py which will resize the images to size (28, 28). The new resized dataset will be located by default in data_split:

python build_dataset.py

• build_dataset.py seems to be so slow, make code more efficient
• probably bug in test/val split
• make better train.py