Quickstart Guide

Installation

Install XFlow using pip:

pip install xflow-py

Basic Usage

XFlow provides a simple and intuitive API for building machine learning pipelines:

  1. Data Pipeline

    from xflow import BasePipeline, InMemoryPipeline

# Create a basic pipeline
pipeline = BasePipeline()

# Or use in-memory pipeline for small datasets
data_pipeline = InMemoryPipeline(data)

  2. Model Creation

    from xflow import BaseModel

# Create a model
model = BaseModel()

  3. Training

    from xflow import BaseTrainer

# Create and configure trainer
trainer = BaseTrainer(model=model, data=pipeline)

# Start training
trainer.train()

  4. Data Transforms (PyTorch Support)

    XFlow now supports PyTorch/torchvision transforms alongside TensorFlow transforms:

    from xflow.data.transform import TransformRegistry, build_transforms_from_config

# Use individual PyTorch transforms
to_tensor = TransformRegistry.get("torch_to_tensor")
resize = TransformRegistry.get("torch_resize")
normalize = TransformRegistry.get("torch_normalize")

# Or build from configuration
config = [
    {"name": "torch_to_tensor"},
    {"name": "torch_resize", "params": {"size": [224, 224]}},
    {"name": "torch_normalize", "params": {"mean": [0.485, 0.456, 0.406], "std": [0.229, 0.224, 0.225]}}
]
transforms = build_transforms_from_config(config)

# Convert pipeline to PyTorch dataset
torch_dataset = pipeline.to_framework_dataset("pytorch")

  5. Training Callbacks (PyTorch Support)

    XFlow now supports PyTorch callbacks for training monitoring and control:

    from xflow.trainers.callback import CallbackRegistry, build_callbacks_from_config

# Use individual PyTorch callbacks
early_stopping = CallbackRegistry.get_handler("torch_early_stopping")(
    monitor="val_loss", patience=10, restore_best=True
)

model_checkpoint = CallbackRegistry.get_handler("torch_model_checkpoint")(
    filepath="best_model.pth", monitor="val_loss", save_best_only=True
)

# Or build from configuration
callback_config = [
    {"name": "torch_early_stopping", "params": {"patience": 10}},
    {"name": "torch_progress_bar", "params": {"desc": "Training"}},
    {"name": "torch_lr_scheduler", "params": {"scheduler_class": "StepLR", "step_size": 10}}
]
callbacks = build_callbacks_from_config(callback_config, framework="torch")

  6. Configuration Management

    from xflow import ConfigManager

# Load configuration
config = ConfigManager.load_config('config.yaml')

# Access configuration values
learning_rate = config.training.learning_rate

Next Steps