Experiment Tracking

Search…
Experiment Tracking
Track and visualize experiments in real time, compare baselines, and iterate quickly on ML projects
Use the wandb Python library to track machine learning experiments with a few lines of code. If you're using a popular framework like PyTorch or Keras, we have lightweight integrations.
You can then review the results in an interactive dashboard or export your data to Python for programmatic access using our Public API.
Integrating W&B in your script
Below are the simple building blocks to track an experiment with W&B. We also have a whole host of special integrations for PyTorch, Keras, Scikit, etc. See our Integration guides.
1.​wandb.init(): Initialize a new run at the top of your script. This returns a Run object and creates a local directory where all logs and files are saved, then streamed asynchronously to a W&B server. If you want to use a private server instead of our hosted cloud server, we offer Self-Hosting.
2.​wandb.config: Save a dictionary of hyperparameters such as learning rate or model type. The model settings you capture in config are useful later to organize and query your results.
3.​wandb.log(): Log metrics over time in a training loop, such as accuracy and loss. By default, when you call wandb.log it appends a new step to the history object and updates the summary object.
history: An array of dictionary-like objects that tracks metrics over time. These time series values are shown as default line plots in the UI.
summary: By default, the final value of a metric logged with wandb.log(). You can set the summary for a metric manually to capture the highest accuracy or lowest loss instead of the final value. These values are used in the table, and plots that compare runs — for example, you could visualize at the final accuracy for all runs in your project.
4.​wandb.log_artifact: Save outputs of a run, like the model weights or a table of predictions. This lets you track not just model training, but all the pipeline steps that affect the final model.
Best Practices
The wandb library is incredibly flexible. Here are some suggested guidelines.
1.Config: Track hyperparameters, architecture, dataset, and anything else you'd like to use to reproduce your model. These will show up in columns— use config columns to group, sort, and filter runs dynamically in the app.
2.Project: A project is a set of experiments you can compare together. Each project gets a dedicated dashboard page, and you can easily turn on and off different groups of runs to compare different model versions.
3.Notes: A quick commit message to yourself, the note can be set from your script and is editable in the table.
4.Tags: Identify baseline runs and favorite runs. You can filter runs using tags, and they're editable in the table.
1
import wandb
2
​
3
config = dict (
4
  learning_rate = 0.01,
5
  momentum = 0.2,
6
  architecture = "CNN",
7
  dataset_id = "peds-0192",
8
  infra = "AWS",
9
)
10
​
11
wandb.init(
12
  project="detect-pedestrians",
13
  notes="tweak baseline",
14
  tags=["baseline", "paper1"],
15
  config=config,
16
)
Copied!
What data is logged?
All the data logged from your script is saved locally to your machine in a directory called wandb, then synced to the W&B cloud or your private server.
Logged Automatically
System metrics: CPU and GPU utilization, network, etc. These are shown in the System tab on the run page. For the GPU, these are fetched with nvidia-smi.
Command line: The stdout and stderr are picked up and show in the logs tab on the run page.​
Turn on Code Saving in your account's Settings page to get:
Git commit: Pick up the latest git commit and see it on the overview tab of the run page, as well as a diff.patch file if there are any uncommitted changes.
Dependencies: The requirements.txt file will be uploaded and shown on the files tab of the run page, along with any files you save to the wandb directory for the run.
Logged with Specific Calls
Where data and model metrics are concerned, you get to decide exactly what you want to log.
Dataset: You have to specifically log images or other dataset samples for them to stream to W&B.
PyTorch gradients: Add wandb.watch(model) to see gradients of the weights as histograms in the UI.
Configuration info: Log hyperparameters, a link to your dataset, or the name of the architecture you're using as config parameters, passed in like this: wandb.init(config=your_config_dictionary).
Metrics: Use wandb.log to see metrics from your model. If you log metrics like accuracy and loss from inside your training loop, you'll get live updating graphs in the UI.
Guides
Launch Experiments with wandb.init
Last modified 8mo ago
Copy link
Contents
Integrating W&B in your script
Best Practices
What data is logged?
Logged Automatically
Logged with Specific Calls