Documentation Index
Fetch the complete documentation index at: https://mintlify.com/MilesONerd/neurenix/llms.txt
Use this file to discover all available pages before exploring further.
The predict command loads a trained model and generates predictions on new input data, with support for various output formats.
Usage
neurenix predict --model <model_file> --input <data_path> [options]
Options
| Option | Type | Default | Description |
|---|
--model | string | required | Path to the trained model file |
--input | string | required | Input data file or directory |
--output | string | predictions.csv | Output file for predictions |
--batch-size | integer | 32 | Batch size for prediction |
--device | string | auto | Device to use (cpu, cuda, auto) |
--format | string | auto | Output format (csv, json, npy) |
- csv: Comma-separated values (default)
- json: JSON array format
- npy: NumPy binary format (requires NumPy)
If --format is not specified, the format is inferred from the output file extension.
Examples
Basic prediction
neurenix predict --model models/classifier.nrx --input data/test.csv
Loading model from models/classifier.nrx...
Loading input data from data/test.csv...
Making predictions with batch size 32...
Saving predictions to predictions.csv...
Predictions saved to predictions.csv
Predict with custom output
neurenix predict \
--model models/model.nrx \
--input data/new_data.csv \
--output results/predictions.csv
Loading model from models/model.nrx...
Loading input data from data/new_data.csv...
Making predictions with batch size 32...
Saving predictions to results/predictions.csv...
Predictions saved to results/predictions.csv
neurenix predict \
--model models/classifier.nrx \
--input data/test.csv \
--output predictions.json \
--format json
Loading model from models/classifier.nrx...
Loading input data from data/test.csv...
Making predictions with batch size 32...
Saving predictions to predictions.json...
Predictions saved to predictions.json
Large batch size for GPU
neurenix predict \
--model models/model.nrx \
--input data/large_dataset.csv \
--batch-size 256 \
--device cuda
Loading model from models/model.nrx...
Loading input data from data/large_dataset.csv...
Making predictions with batch size 256...
Saving predictions to predictions.csv...
Predictions saved to predictions.csv
neurenix predict \
--model models/model.nrx \
--input data/images/ \
--output predictions.npy \
--format npy
Loading model from models/model.nrx...
Loading input data from data/images/...
Making predictions with batch size 32...
Saving predictions to predictions.npy...
Predictions saved to predictions.npy
CPU-only prediction
neurenix predict \
--model models/model.nrx \
--input data/test.csv \
--device cpu
--input parameter accepts:
- Single file: CSV, JSON, or other supported formats
- Directory: Loads all compatible files in the directory
# Single file
neurenix predict --model model.nrx --input test_data.csv
# Directory of images
neurenix predict --model model.nrx --input images/test/
Error Handling
Model not found
neurenix predict --model missing.nrx --input data.csv
Error: Model file 'missing.nrx' not found.
neurenix predict --model model.nrx --input missing.csv
Error: Input 'missing.csv' not found.
Prediction error
neurenix predict --model model.nrx --input incompatible_data.csv
Loading model from model.nrx...
Loading input data from incompatible_data.csv...
Error making predictions: Input shape mismatch
Batch Processing
For large datasets, adjust batch size based on available memory:
# Small batch for CPU
neurenix predict --model model.nrx --input large.csv --batch-size 16 --device cpu
# Large batch for GPU
neurenix predict --model model.nrx --input large.csv --batch-size 512 --device cuda
Performance Tip: Larger batch sizes generally provide better throughput on GPUs, while smaller batches may be necessary for CPU or limited memory scenarios.
Prediction Pipeline
1. Load trained model
neurenix predict --model models/production_model.nrx --input new_data.csv
2. Make predictions
The model processes input data in batches and generates predictions.
3. Save results
Predictions are saved in the specified format and location.
Best Practices
1. Use appropriate batch sizes
Match batch size to your hardware capabilities:
# CPU: smaller batches
neurenix predict --model model.nrx --input data.csv --batch-size 32 --device cpu
# GPU: larger batches
neurenix predict --model model.nrx --input data.csv --batch-size 256 --device cuda
# CSV for tabular data
neurenix predict --model model.nrx --input data.csv --format csv
# JSON for structured output
neurenix predict --model model.nrx --input data.csv --format json
# NPY for numerical arrays
neurenix predict --model model.nrx --input data.csv --format npy
3. Organize prediction outputs
Store predictions in organized directories:
mkdir -p predictions/$(date +%Y%m%d)
neurenix predict \
--model models/model.nrx \
--input data/batch_1.csv \
--output predictions/$(date +%Y%m%d)/batch_1_predictions.csv
4. Use auto device selection
Let Neurenix choose the best available device:
neurenix predict --model model.nrx --input data.csv --device auto
Integration Examples
Batch prediction script
#!/bin/bash
for file in data/test_batch_*.csv; do
output="predictions/$(basename $file .csv)_pred.csv"
neurenix predict --model models/best.nrx --input $file --output $output
done
Python integration
import subprocess
import os
def run_predictions(model_path, input_data, output_path):
cmd = [
"neurenix", "predict",
"--model", model_path,
"--input", input_data,
"--output", output_path,
"--device", "auto"
]
subprocess.run(cmd, check=True)
run_predictions(
"models/classifier.nrx",
"data/test.csv",
"results/predictions.csv"
)
See Also