Experiments
Track your machine learning experiments with detailed logging and comparison tools.
Creating an Experiment
Navigate to Deep Learning Platform → Experiments → Click Create
Basic Information
Experiment Name* (Required)
Enter a descriptive name for the experiment
Example:
image-classification-resnet,nlp-sentiment-bert
Description (Optional)
Detailed description of experiment purpose and goals
Framework* (Required)
Select your ML framework from dropdown:
PyTorch
TensorFlow
Scikit-learn
Keras
Others
Default:
pytorch
Task Type* (Required)
Select the ML task type:
Classification
Regression
Detection
Segmentation
Others
Default:
classification
Model Type* (Required)
Specify the model architecture
Examples: ResNet, BERT, YOLO, Custom
Project ID (Optional)
Link experiment to a specific project
Training Configuration
Epochs* (Required)
Number of training epochs
Example:
100
Number of training epochs (Helper text)
Additional context for epochs setting
Staging batch size (Optional)
Batch size for staging/validation
Learning Rate* (Required)
Initial learning rate for training
Example:
0.001Helper text: "Valid learning rate"
Loss Function* (Required)
Select optimizer from dropdown:
Adam
SGD
RMSprop
Others
Default:
adam
Loss Function* (Required)
Select loss function:
Categorical Crossentropy
Binary Crossentropy
MSE
Others
Default:
categorical_crossentropy
Environment & Resources
Python Version* (Required)
Select Python version:
Python 3.9
Python 3.10
Python 3.11
Default:
python
GPU Required (Checkbox)
Check if GPU is required for training
Memory Requirement (GB)* (Required)
Required memory in GB
Example:
8,16,32
Required memory in GB (Helper text)
CPU Cores* (Required)
Number of CPU cores needed
Example:
4,8,16
Number of CPU cores (Helper text)
Metadata
Tags (Optional)
Comma-separated tags for organizing experiments
Example:
computer-vision, production, baseline
Notes (Optional)
Additional notes or comments about the experiment
Public Experiment (Checkbox)
Make experiment visible to all organization members
Actions
Cancel: Discard and close the form
Create Experiment: Submit and create the experiment
Example Configuration
Viewing Experiment Details
To view detailed information about an experiment:
Navigate to Deep Learning Platform → Experiments
Click on an experiment from the list
View comprehensive details in the modal dialog
Details Panel Sections:
Basic Information:
Experiment Name: e.g., "Image Classification CNN"
Description: Full description of the experiment
Framework: TensorFlow, PyTorch, etc.
Task Type: Classification, Regression, etc.
Model Type: CNN, ResNet, Custom, etc.
Project ID: Associated project
Training Configuration:
Epochs: Number of training epochs (e.g., 100)
Batch Size: Training batch size (e.g., 32)
Learning Rate: Initial learning rate (e.g., 0.001)
Optimizer: Adam, SGD, etc.
Loss Function: Categorical Crossentropy, MSE, etc.
Environment & Resources:
Python Version: e.g., Python 3.9
GPU Required: Checkbox status
Memory Requirement (GB): e.g., 8 GB
CPU Cores: e.g., 4 cores
Metadata:
Tags: Comma-separated tags
Notes: Additional notes
Public Experiment: Visibility status
Creator and timestamps
Editing an Experiment
To modify an experiment configuration:
Navigate to the experiment details page
Click Edit button (or three-dot menu → Edit)
Modify editable fields in the Edit Experiment modal
Click Update Experiment to save changes
[!NOTE] The Edit form looks very similar to the View form, but fields become editable and you'll see an "Update Experiment" button instead of just "Cancel".
[!NOTE] You cannot edit core configuration (framework, resources, hyperparameters) of a running or completed experiment. To try different settings, clone the experiment instead.
Editable Fields:
✅ Description
✅ Tags
✅ Notes
✅ Public/Private status
❌ Framework (cannot edit)
❌ Resources (cannot edit while running)
❌ Hyperparameters (cannot edit)
Cloning an Experiment
To create a copy of an experiment with modified settings:
Open experiment details
Click Clone button
Modify configuration as needed
Give it a new name
Click Create Experiment
Use Cases:
Try different hyperparameters
Run with more/less resources
Test on different datasets
Reproduce results
Deleting an Experiment
To remove an experiment:
Navigate to experiment details or list
Click Delete button (trash icon)
Confirm deletion in the dialog
Experiment and associated data will be removed
[!WARNING] Deleting an experiment will permanently remove:
Experiment configuration
Training logs
Metrics and charts
Saved checkpoints (unless linked to a registered model)
This action cannot be undone!
Before Deleting:
Export important logs or metrics
Register any valuable models
Download artifacts if needed
Verify you have the correct experiment selected
Monitoring Experiments
Once submitted, track your experiment:
Real-time Monitoring
View live logs
Monitor resource utilization (CPU, GPU, memory)
Track metrics as they're logged
Receive alerts on failures
Experiment Status
Pending: Waiting for resources
Running: Currently executing
Completed: Finished successfully
Failed: Encountered an error
Stopped: Manually stopped
Cancelled: Cancelled before starting
Actions Available
View Logs: See stdout/stderr
View Metrics: Charts and graphs
Stop: Terminate running experiment
Clone: Create a copy with same config
Compare: Compare with other experiments
Export: Download results and artifacts
Comparing Experiments
Compare multiple experiments side-by-side:
Select Experiments: Check boxes for 2+ experiments
Click Compare: Opens comparison view
View Differences:
Hyperparameters table
Metrics charts (overlaid)
Resource usage comparison
Final results summary
Best Practices
Naming Conventions
Tagging Strategy
Domain:
computer-vision,nlp,audioTask:
classification,detection,segmentationStage:
exploration,tuning,production
Resource Optimization
Start with minimal resources, scale up as needed
Use GPU only when necessary
Monitor resource utilization
Next Steps
Register your trained model in Models
Deploy to production via Deployments
Monitor performance in Analytics
Last updated