Documentation Index
Fetch the complete documentation index at: https://docs.onyx-robotics.com/llms.txt
Use this file to discover all available pages before exploring further.
This reference documents all public functions and classes in the onyxengine Python package.
Installation
Quick Reference
Core Functions
| Function | Description |
|---|
save_dataset() | Upload a dataset to the Engine |
load_dataset() | Download a dataset from the Engine |
save_model() | Upload a model to the Engine |
load_model() | Download a model from the Engine |
train_model() | Train a model on the Engine |
optimize_model() | Run hyperparameter optimization |
Data Classes
| Class | Description |
|---|
OnyxDataset | Dataset container with dataframe and metadata |
Model Configurations
| Class | Description |
|---|
MLPConfig | Multi-Layer Perceptron configuration |
RNNConfig | Recurrent Neural Network configuration |
TransformerConfig | Transformer configuration |
Feature Classes
| Class | Description |
|---|
Input | Model input feature definition |
Output | Model output feature definition |
Training Classes
| Class | Description |
|---|
TrainingConfig | Training parameters |
OptimizationConfig | Hyperparameter search configuration |
| Optimizer Configs | AdamW, SGD, and scheduler configurations |
Import Patterns
# Initialize the client
from onyxengine import Onyx
onyx = Onyx() # Uses ONYX_API_KEY env var
# Or explicitly: onyx = Onyx(api_key="your_api_key")
# Client methods
onyx.save_dataset(...)
onyx.load_dataset(...)
onyx.save_model(...)
onyx.load_model(...)
onyx.train_model(...)
onyx.optimize_model(...)
onyx.get_object_metadata(...)
# Data classes
from onyxengine.data import OnyxDataset
# Modeling classes
from onyxengine.modeling import (
# Feature classes
Input,
Output,
# Model configs
MLPConfig,
RNNConfig,
TransformerConfig,
# Training configs
TrainingConfig,
OptimizationConfig,
# Optimizers
AdamWConfig,
SGDConfig,
# Learning rate schedulers
CosineDecayWithWarmupConfig,
CosineAnnealingWarmRestartsConfig,
# Optimization configs (for hyperparameter search)
MLPOptConfig,
RNNOptConfig,
TransformerOptConfig,
AdamWOptConfig,
SGDOptConfig,
CosineDecayWithWarmupOptConfig,
CosineAnnealingWarmRestartsOptConfig,
)
Type Conventions
The SDK uses Python type hints throughout:
from typing import List, Dict, Optional, Union, Literal
# Example function signature
def load_model(
name: str,
version_id: Optional[str] = None,
mode: Literal["online", "offline"] = "online"
) -> Union[MLP, RNN, Transformer]:
...
Error Handling
Client methods raise exceptions with descriptive messages:
from onyxengine import Onyx
onyx = Onyx()
try:
model = onyx.load_model('nonexistent_model')
except Exception as e:
print(f"Error: {e}")
# "Onyx Engine API error: Model [nonexistent_model: None] not found in the Engine."
Environment Variables
| Variable | Description |
|---|
ONYX_API_KEY | Your API key for authentication |
Local Storage
The SDK caches datasets and models locally:
- Datasets:
~/.onyx/datasets/
- Models:
~/.onyx/models/
Cached files are automatically updated when newer versions are available.
