Hey everyone! Just spent the afternoon building a really clean model comparison dashboard using Weights & Biases Tables, and I'm pretty happy with how it turned out. I know a lot of folks here are evaluating different experiment trackers, so I figured I'd share my setup and a few screenshots.
I was comparing three different architectures for an image classification task. The pain point was always having to juggle between different runs to see which model performed best on which metrics across validation folds. W&B Tables ended up being the perfect tool for this.
Here's the core snippet I used to log the comparison table:
```python
import wandb
# After validation loop for each model
table = wandb.Table(columns=["model", "fold", "accuracy", "precision", "recall", "f1", "inference_time"])
for fold in range(num_folds):
table.add_data(model_name, fold, acc, prec, rec, f1, inf_time)
wandb.log({"model_comparison": table})
```
Once all my runs were logged, I created a new dashboard panel and grouped the tables from each run. The beauty is you can easily:
* Sort by any column (quickly found the model with the highest F1)
* Filter out specific folds
* Visually highlight cells based on thresholds (e.g., all accuracy > 0.9 in green)
* Export the entire table to CSV for further analysis
I've attached a screenshot of the final dashboard. You can see at a glance that "Model B" had the best balance of accuracy and inference speed across all folds. This is so much clearer than flipping through individual run pages or trying to build this in a spreadsheet.
For anyone starting out, I'd definitely recommend playing with Tables for any kind of head-to-head model evaluation. It's a game-changer compared to just logging individual metrics. Plus, you can embed plots next to the tables in the same dashboard for a holistic view.
What's your go-to method for model comparisons? Anyone else using W&B Tables for something similar?
Dashboards or it didn't happen.