Getting Started with ns&i requires a systematic approach to environment setup, data preparation, and model initialization. This guide provides the exact checklist you need to move from installation to your first successful training run. It is designed for beginners and general enthusiasts who are new to this specific neural architecture and want a clear, low-friction entry point.
You do not need advanced mathematics or prior research experience to begin. You do, however, need a compatible Python environment and a basic understanding of file directories. The steps below ensure you avoid common configuration errors that plague new users.
Core Environment Preparation
Before touching the code, you must verify your system compatibility. The ns&i library relies heavily on specific tensor operations that require modern hardware acceleration. Ensure your Python version is current and that your GPU drivers are up to date.
Use a virtual environment to isolate dependencies. This prevents conflicts with other machine learning tools on your machine. Create a directory for your project and initialize a git repository to track your file changes from day one.
- Install the core package and its dependencies.
- Verify GPU visibility using your hardware monitoring tool.
- Set up a clean virtual environment for isolated development.
- Create the root project directory and subfolders for data.
Configuring the Project Structure
ns&i expects a strict file hierarchy to function correctly. Deviating from the expected structure causes silent failures during data loading. Your project root should contain a config directory and a data directory at minimum.
The configuration files handle the hyperparameters and training schedules. You will edit these files frequently as you adjust your experiments. Keep the default settings unchanged for your first run to establish a working baseline.
- config.yaml: Defines the model architecture and optimizer settings.
- data/train.csv: Contains the training feature set.
- data/val.csv: Contains the validation feature set.
- output/: An empty directory where trained weights will be saved.
Running the First Training Loop
With the structure in place, you can execute the initial training script. This step validates that your data pipeline connects properly to the model. Do not run large datasets on your first attempt. Use a small sample to confirm the loop terminates correctly.
Monitor the loss values printed to the console. A sharp drop indicates the model is learning from your sample data. If the loss remains flat or errors occur, check your data formats immediately. The Getting Started with ns&i: The Ultimate Beginner Checklist ends here with a working prototype.
Validation and Next Steps
Once the first run completes, inspect the files in the output directory. Ensure the saved model artifacts are readable and non-zero in size. This confirms that your save logic functions as intended.
From here, you can begin scaling your dataset. Start by adding more features to your input file. Then adjust the learning rate in the config file to observe different convergence behaviors. Incremental changes allow you to isolate the cause of any new errors.
| Checklist Item | Expected Outcome | Common Error |
|---|---|---|
| Install Dependencies | Package versions match the documentation | Python version mismatch |
| Directory Setup | All subfolders exist in the root | Case-sensitivity path errors |
| Small Sample Run | Loss decreases and finishes | Shape mismatch in tensors |
Frequently Asked Questions
Do I need a GPU to run the initial setup?
No, a CPU is sufficient for the initial baseline run. A GPU is required only when you scale the dataset to full size for serious training.
What happens if the loss does not decrease?
If the loss remains flat, check your learning rate. A value that is too high causes divergence, while a value that is too low causes stagnation. Adjust the config file and rerun the sample.
Can I use my own data format?
You can use your own data, but it must be converted to the standard CSV format expected by the loader. Ensure that your feature columns align perfectly with the model input dimensions.