wandb

🪐 spaCy Project: Weights & Biases integration

⚠️ This project template uses the new spaCy v3.0, which is currently available as a nightly pre-release. You can install it from pip as spacy-nightly: pip install spacy-nightly. Make sure to use a fresh virtual environment.

Use Weights & Biases for logging of training experiments. This project template uses the IMDB Movie Review Dataset and includes two workflows: log for training a simple text classification model and logging the results to Weights & Biases (works out-of-the-box and only requires the [training.logger] to be set in the config) and parameter-search for programmatically creating variants of the config for a simple hyperparameter grid search, running the experiments and logging the results.

📋 project.yml

The project.yml defines the data assets required by the project, as well as the available commands and workflows. For details, see the spaCy projects documentation.

⏯ Commands

The following commands are defined by the project. They can be executed using spacy project run [name]. Commands are only re-run if their inputs have changed.

Command	Description
install	Install dependencies and log in to Weights & Biases
data	Extract the gold-standard annotations
train	Train a model using the default config
configs-search	Create variations of the initial, default file for IMDB sentiment classification using different combinations of hyperparameters
train-search	Run customized training runs for hyperparameter search using the created configs

⏭ Workflows

The following workflows are defined by the project. They can be executed using spacy project run [name] and will run the specified commands in order. Commands are only re-run if their inputs have changed.

Workflow	Steps
log	data → train
parameter-search	data → configs-search → train-search

🗂 Assets

The following assets are defined by the project. They can be fetched by running spacy project assets in the project directory.

File	Source	Description
assets/aclImdb_v1.tar.gz	URL	Movie Review Dataset for sentiment analysis by Maas et al., ACL 2011.