Handwriting Digit Recognizer using Deep Learning

---

• Inspiration: Kaggle Digit Recognizer

Installation

From the source directory run the following commands

Virtual Env Creation & Activation

• python3 -m venv venv for initialising the virtual environment
• source venv/bin/activate for activating the virtual environment

Dependency Installation

The following commands shall be ran after activating the virtual environment.

• pip install --upgrade pip for upgrading the pip
• pip install -r requirements.txt for the functional dependencies
• pip install -r requirements-dev.txt for the development dependencies. (should include pre-commit module)
• pre-commit install for installing the precommit hook

For the extra modules, which are not a standard pip modules (either from your own src or from any github repo)

• pip install -e . for the files/modules in src to be accessed as a package. This is accompanied with setup.py and setup.cfg files
  • -e means installing a project in editable mode, thus any local modifications made to the code will take effect without reinstallation.

Result Evaluation

Model	On the Validation set	On Kaggle set
Basic Conv	100%	99%
Mobilenet	xx.xx%	xx.xx%
VGG16	xx.xx%	xx.xx%
Resnet50	99.00%	98.85%
Resnet18	99.19%	99.17%
WideResnet28-10	xx.xx%	xx.xx%

Pytorch Lightning

• To activate Tensorboard: tensorboard --logdir=model_chkpt/lightning_logs/

Training Methodology

• To train a model: python ./src/train.py -m <model_name> -e <epochs> -bs <batch_size> -lr <learning_rate> -v <version>

Model Architectures

• Mobilenet
• VGG16
• Resnet164
• WideResnet28-10

Ensembling methods

• Unweighted average
• Majority voting
• Super Learner

Pytest

pytest --durations=0 #Show all times for tests and setup and teardown

pytest --durations=1 #Just show me the slowest