Does this skill support both PyTorch and TensorFlow?

Yes, it provides comprehensive implementation patterns for both the SummaryWriter in PyTorch and the Keras TensorBoard callback in TensorFlow.

How do I view the TensorBoard visualizations?

Once your logs are generated, you can launch the dashboard using the command 'tensorboard --logdir=runs' and access the interface through your web browser at http://localhost:6006.

Can I use this for hyperparameter tuning?

Absolutely. The skill includes advanced features for tracking hyperparameters and comparing results across different experiment configurations using the HParams dashboard.

Is it possible to visualize non-scalar data like images?

Yes, the skill supports logging individual images, image grids, and text to inspect model inputs and predictions qualitatively during the training process.

Can it help debug performance issues?

Yes, it includes guidance on performance profiling to help you identify bottlenecks in your training pipeline and optimize hardware utilization.

TensorBoard ML Visualization

Name: TensorBoard ML Visualization
Author: zechenzhangAGI

byzechenzhangAGI

•

384

データサイエンスとML

Visualizes machine learning training metrics and model performance to streamline experiment tracking and model debugging.

概要

TensorBoard is the definitive open-source toolkit for machine learning experimentation, providing robust visualization capabilities for PyTorch and TensorFlow workflows. This skill empowers Claude to help you implement real-time tracking for loss and accuracy, debug complex model architectures with interactive graphs, and visualize high-dimensional embeddings. Whether you are tuning hyperparameters, analyzing gradient distributions, or profiling performance bottlenecks, this skill provides the standardized implementation patterns and best practices required to transform raw training logs into actionable research insights.

主な機能

Real-time visualization of scalars like loss, accuracy, and learning rates
Comprehensive hyperparameter tuning comparison across multiple runs
High-dimensional data projection using PCA, t-SNE, and UMAP
Performance profiling to identify and resolve training bottlenecks
Interactive model graph exploration to debug architecture and data flow
384 GitHub stars

ユースケース

Debugging vanishing or exploding gradients using weight histograms
Visualizing model predictions and sample inputs to verify data quality
Comparing training runs across different learning rates and architectures

概要

主な機能

Real-time visualization of scalars like loss, accuracy, and learning rates
Comprehensive hyperparameter tuning comparison across multiple runs
High-dimensional data projection using PCA, t-SNE, and UMAP
Performance profiling to identify and resolve training bottlenecks
Interactive model graph exploration to debug architecture and data flow
384 GitHub stars

ユースケース

Debugging vanishing or exploding gradients using weight histograms
Visualizing model predictions and sample inputs to verify data quality
Comparing training runs across different learning rates and architectures