Skip to main content

AI Trainer

Project description

The Mission

There are still tasks that can be faster explained to a human than to a computer. This project came to life for the purpose of solving that.

Using trainer you can train artificial brains by stating a set of inputs and either your desired output or a combination of output and easily to define constraints.

Take for example the following traditional medical multi-class classification problem: Given one or more ultrasound videos, an xray and a textual description of the patient's symptoms, predict the medical decision of care. From an engineering perspective, this is a difficult task. With trainer tasks that involve decisions given numerous different inputs should be greatly simplified.

Installation for User

Open anaconda powershell, activate an environment with anaconda, navigate into the trainer repo and execute the following to install trainer using pip, including its dependencies:

pip install ai-trainer

For Online Learning you have to install PyTorch:

conda install pytorch torchvision cudatoolkit=10.1 -c pytorch

AI-Trainer helps with building a data generator and it relies on imgaug for it:

conda install imgaug -c conda-forge

Getting started with training models

Trainer currently supports annotating images and videos. First, create a dataset using

trainer init-ds
cd YOUR_DATASET

Getting started with using trainer in python

For using the annotated data, you can use trainer as a python package. After activating the environment containing the trainer and its dependencies, feel free to inspect some of the tutorials in ./tutorials/.

Development Setup

Execute the user installation, but instead of using pip install ai-trainer, clone the repo locally.

git clone https://github.com/Telcrome/ai-trainer

Both vsc and pycharm are used for development with their configurations provided in .vscode and .idea

Recommended environments

For development we recommend to install the conda environment into a subfolder of the repo. This allows for easier experimentation and the IDE expects it this way.

conda env create --prefix ./envs -f environment.yml
conda activate .\envs\.

Now install a deep learning backend. PyTorch provides well-working conda install commands.

For Tensorflow with GPU:

conda install cudatoolkit=10.0 cudnn=7.6.0=cuda10.0_0
pip install tensorflow-gpu

Testing Development for pip and cli tools

Installing the folder directly using pip does not work due to the large amount of files inside the local development folder, especially because in the local development setup the environment is expected to be a subfolder of the repo.

pip install -e .

Using Docker

Docker and the provided DOCKERFILE support is currently experimental as it proved to slow down the annotation GUI too much. When the transition to a web GUI is completed docker will be supported again.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for ai-trainer, version 0.0.3
Filename, size File type Python version Upload date Hashes
Filename, size ai-trainer-0.0.3.tar.gz (40.2 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page