​spaCy is a popular "industrial-strength" NLP library: fast, accurate models with a minimum of fuss. As of spaCy v3, Weights and Biases can now be used with spacy train to track your spaCy model's training metrics as well as to save and version your models and datasets. And all it takes is a few added lines in your configuration!

Getting Started: Track and Save your Models

1) Install the wandb library and log in

!pip install wandb
​
import wandb
wandb.login()

pip install wandb
wandb login

2) Add the WandbLogger to your spaCy config file

spaCy config files are used to specify all aspects of training, not just logging -- GPU allocation, optimizer choice, dataset paths, and more. Minimally, under [training.logger] you need to provide the key @loggers with the value "spacy.WandbLogger.v2", plus a project_name. You can also turn on dataset and model versioning by just adding a line to the config file.

[training.logger]
@loggers = "spacy.WandbLogger.v2"
project_name = "my_spacy_project"
remove_config_values = ["paths.train", "paths.dev", "corpora.train.path", "corpora.dev.path"]
log_dataset_dir = "./corpus"
model_log_interval = 1000

3) Start training

Once you have added the WandbLogger to your spaCy training config you can run spacy train as usual.

!python -m spacy train \
    config.cfg \
    --output ./output \
    --paths.train ./train \
    --paths.dev ./dev

python -m spacy train \
    config.cfg \
    --output ./output \
    --paths.train ./train \
    --paths.dev ./dev

When training begins, a link to your training run's W&B page will be output which will take you to this run's experiment tracking dashboard in the Weights & Biases web UI.

Data Visualization with W&B Tables

Use W&B Tables to log, query, and analyze your data. You can think of a W&B Table as a DataFrame that you can interact with inside W&B. Tables support rich media types, primitive and numeric types, as well as nested lists and dictionaries.

This pseudo-code shows you how to log images, along with their ground truth and predicted class, to W&B Tables. You can use the same structure to log your text data:

# Create a new W&B Run
wandb.init(project="mnist")
​
# Create a W&B Table
my_table = wandb.Table(columns=["id", "image", "labels", "prediction"])
​
# Get your image data and make predictions
image_tensors, labels = get_mnist_data()
predictions = model(image_tensors)
​
# Add your image data and predictions to the W&B Table
for idx, im in enumerate(image_tensors): 
  my_table.add_data(idx, wandb.Image(im), labels[idx], predictions[id])
​
# Log your Table to W&B
wandb.log({"mnist_predictions": my_table})

This is will produce a Table like this:

For more examples of data visualization with W&B Tables, please see the documentation. ​