Getting Started — Ray 3.0.0.dev0

Data: Scalable Datasets for ML

Scale offline inference and training ingest with Ray Data – a data processing library designed for ML.

To learn more, see Offline batch inference and Data preprocessing and ingest for ML training.

Note

To run this example, install Ray Data:

pip install -U "ray[data]"

from typing import Dict
import numpy as np
import ray

# Create datasets from on-disk files, Python objects, and cloud storage like S3.
ds = ray.data.read_csv("s3://anonymous@ray-example-data/iris.csv")

# Apply functions to transform data. Ray Data executes transformations in parallel.
def compute_area(batch: Dict[str, np.ndarray]) -> Dict[str, np.ndarray]:
    length = batch["petal length (cm)"]
    width = batch["petal width (cm)"]
    batch["petal area (cm^2)"] = length * width
    return batch

transformed_ds = ds.map_batches(compute_area)

# Iterate over batches of data.
for batch in transformed_ds.iter_batches(batch_size=4):
    print(batch)

# Save dataset contents to on-disk files or cloud storage.
transformed_ds.write_parquet("local:///tmp/iris/")

Learn more about Ray Data

Train: Distributed Model Training

Ray Train abstracts away the complexity of setting up a distributed training system.

PyTorch

This example shows how you can use Ray Train with PyTorch.

To run this example install Ray Train and PyTorch packages:

Note

pip install -U "ray[train]" torch torchvision

Set up your dataset and model.

import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.transforms import ToTensor

def get_dataset():
    return datasets.FashionMNIST(
        root="/tmp/data",
        train=True,
        download=True,
        transform=ToTensor(),
    )

class NeuralNetwork(nn.Module):
    def __init__(self):
        super().__init__()
        self.flatten = nn.Flatten()
        self.linear_relu_stack = nn.Sequential(
            nn.Linear(28 * 28, 512),
            nn.ReLU(),
            nn.Linear(512, 512),
            nn.ReLU(),
            nn.Linear(512, 10),
        )

    def forward(self, inputs):
        inputs = self.flatten(inputs)
        logits = self.linear_relu_stack(inputs)
        return logits

Now define your single-worker PyTorch training function.

def train_func():
    num_epochs = 3
    batch_size = 64

    dataset = get_dataset()
    dataloader = DataLoader(dataset, batch_size=batch_size)

    model = NeuralNetwork()

    criterion = nn.CrossEntropyLoss()
    optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

    for epoch in range(num_epochs):
        for inputs, labels in dataloader:
            optimizer.zero_grad()
            pred = model(inputs)
            loss = criterion(pred, labels)
            loss.backward()
            optimizer.step()
        print(f"epoch: {epoch}, loss: {loss.item()}")

This training function can be executed with:

train_func()

Convert this to a distributed multi-worker training function.

Use the ray.train.torch.prepare_model and ray.train.torch.prepare_data_loader utility functions to set up your model and data for distributed training. This automatically wraps the model with DistributedDataParallel and places it on the right device, and adds DistributedSampler to the DataLoaders.

import ray.train.torch

def train_func_distributed():
    num_epochs = 3
    batch_size = 64

    dataset = get_dataset()
    dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=True)
    dataloader = ray.train.torch.prepare_data_loader(dataloader)

    model = NeuralNetwork()
    model = ray.train.torch.prepare_model(model)

    criterion = nn.CrossEntropyLoss()
    optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

    for epoch in range(num_epochs):
        if ray.train.get_context().get_world_size() > 1:
            dataloader.sampler.set_epoch(epoch)

        for inputs, labels in dataloader:
            optimizer.zero_grad()
            pred = model(inputs)
            loss = criterion(pred, labels)
            loss.backward()
            optimizer.step()
        print(f"epoch: {epoch}, loss: {loss.item()}")

Instantiate a TorchTrainer with 4 workers, and use it to run the new training function.

from ray.train.torch import TorchTrainer
from ray.train import ScalingConfig

# For GPU Training, set `use_gpu` to True.
use_gpu = False

trainer = TorchTrainer(
    train_func_distributed,
    scaling_config=ScalingConfig(num_workers=4, use_gpu=use_gpu)
)

results = trainer.fit()

TensorFlow

This example shows how you can use Ray Train to set up Multi-worker training with Keras.

To run this example install Ray Train and Tensorflow packages:

Note

pip install -U "ray[train]" tensorflow

Set up your dataset and model.

import sys
import numpy as np

if sys.version_info >= (3, 12):
    # Tensorflow is not installed for Python 3.12 because of keras compatibility.
    sys.exit(0)
else:
    import tensorflow as tf

def mnist_dataset(batch_size):
    (x_train, y_train), _ = tf.keras.datasets.mnist.load_data()
    # The `x` arrays are in uint8 and have values in the [0, 255] range.
    # You need to convert them to float32 with values in the [0, 1] range.
    x_train = x_train / np.float32(255)
    y_train = y_train.astype(np.int64)
    train_dataset = tf.data.Dataset.from_tensor_slices(
        (x_train, y_train)).shuffle(60000).repeat().batch(batch_size)
    return train_dataset


def build_and_compile_cnn_model():
    model = tf.keras.Sequential([
        tf.keras.layers.InputLayer(input_shape=(28, 28)),
        tf.keras.layers.Reshape(target_shape=(28, 28, 1)),
        tf.keras.layers.Conv2D(32, 3, activation='relu'),
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(128, activation='relu'),
        tf.keras.layers.Dense(10)
    ])
    model.compile(
        loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
        optimizer=tf.keras.optimizers.SGD(learning_rate=0.001),
        metrics=['accuracy'])
    return model

Now define your single-worker TensorFlow training function.

def train_func():
    batch_size = 64
    single_worker_dataset = mnist_dataset(batch_size)
    single_worker_model = build_and_compile_cnn_model()
    single_worker_model.fit(single_worker_dataset, epochs=3, steps_per_epoch=70)

This training function can be executed with:

    train_func()

Now convert this to a distributed multi-worker training function.

Set the global batch size - each worker processes the same size batch as in the single-worker code.
Choose your TensorFlow distributed training strategy. This examples uses the MultiWorkerMirroredStrategy.

import json
import os

def train_func_distributed():
    per_worker_batch_size = 64
    # This environment variable will be set by Ray Train.
    tf_config = json.loads(os.environ['TF_CONFIG'])
    num_workers = len(tf_config['cluster']['worker'])

    strategy = tf.distribute.MultiWorkerMirroredStrategy()

    global_batch_size = per_worker_batch_size * num_workers
    multi_worker_dataset = mnist_dataset(global_batch_size)

    with strategy.scope():
        # Model building/compiling need to be within `strategy.scope()`.
        multi_worker_model = build_and_compile_cnn_model()

    multi_worker_model.fit(multi_worker_dataset, epochs=3, steps_per_epoch=70)

Instantiate a TensorflowTrainer with 4 workers, and use it to run the new training function.

    from ray.train.tensorflow import TensorflowTrainer
    from ray.train import ScalingConfig

    # For GPU Training, set `use_gpu` to True.
    use_gpu = False

    trainer = TensorflowTrainer(train_func_distributed, scaling_config=ScalingConfig(num_workers=4, use_gpu=use_gpu))

    trainer.fit()

Learn more about Ray Train

Tune: Hyperparameter Tuning at Scale

Tune is a library for hyperparameter tuning at any scale. With Tune, you can launch a multi-node distributed hyperparameter sweep in less than 10 lines of code. Tune supports any deep learning framework, including PyTorch, TensorFlow, and Keras.

Note

To run this example, install Ray Tune:

pip install -U "ray[tune]"

This example runs a small grid search with an iterative training function.

from ray import train, tune


def objective(config):  # ①
    score = config["a"] ** 2 + config["b"]
    return {"score": score}


search_space = {  # ②
    "a": tune.grid_search([0.001, 0.01, 0.1, 1.0]),
    "b": tune.choice([1, 2, 3]),
}

tuner = tune.Tuner(objective, param_space=search_space)  # ③

results = tuner.fit()
print(results.get_best_result(metric="score", mode="min").config)

If TensorBoard is installed, automatically visualize all trial results:

tensorboard --logdir ~/ray_results

Learn more about Ray Tune

Serve: Scalable Model Serving

Ray Serve is a scalable model-serving library built on Ray.

Note

To run this example, install Ray Serve and scikit-learn:

pip install -U "ray[serve]" scikit-learn

This example runs serves a scikit-learn gradient boosting classifier.

import requests
from starlette.requests import Request
from typing import Dict

from sklearn.datasets import load_iris
from sklearn.ensemble import GradientBoostingClassifier

from ray import serve


# Train model.
iris_dataset = load_iris()
model = GradientBoostingClassifier()
model.fit(iris_dataset["data"], iris_dataset["target"])


@serve.deployment
class BoostingModel:
    def __init__(self, model):
        self.model = model
        self.label_list = iris_dataset["target_names"].tolist()

    async def __call__(self, request: Request) -> Dict:
        payload = (await request.json())["vector"]
        print(f"Received http request with data {payload}")

        prediction = self.model.predict([payload])[0]
        human_name = self.label_list[prediction]
        return {"result": human_name}


# Deploy model.
serve.run(BoostingModel.bind(model), route_prefix="/iris")

# Query it!
sample_request_input = {"vector": [1.2, 1.0, 1.1, 0.9]}
response = requests.get(
    "http://localhost:8000/iris", json=sample_request_input)
print(response.text)

As a result you will see {"result": "versicolor"}.

Learn more about Ray Serve

RLlib: Industry-Grade Reinforcement Learning

RLlib is an industry-grade library for reinforcement learning (RL) built on top of Ray. RLlib offers high scalability and unified APIs for a variety of industry- and research applications.

Note

To run this example, install rllib and either tensorflow or pytorch:

pip install -U "ray[rllib]" tensorflow  # or torch

import gymnasium as gym
import numpy as np
import torch

from ray.rllib.algorithms.ppo import PPOConfig


# Define your problem using python and Farama-Foundation's gymnasium API:
class SimpleCorridor(gym.Env):
    """Corridor in which an agent must learn to move right to reach the exit.

    ---------------------
    | S | 1 | 2 | 3 | G |   S=start; G=goal; corridor_length=5
    ---------------------

    Possible actions to chose from are: 0=left; 1=right
    Observations are floats indicating the current field index, e.g. 0.0 for
    starting position, 1.0 for the field next to the starting position, etc..
    Rewards are -0.1 for all steps, except when reaching the goal (+1.0).
    """

    def __init__(self, config):
        self.end_pos = config["corridor_length"]
        self.cur_pos = 0.0
        self.action_space = gym.spaces.Discrete(2)  # left and right
        self.observation_space = gym.spaces.Box(0.0, self.end_pos, (1,), np.float32)

    def reset(self, *, seed=None, options=None):
        """Resets the episode.

        Returns:
           Initial observation of the new episode and an info dict.
        """
        self.cur_pos = 0.0
        # Return initial observation.
        return np.array([self.cur_pos], np.float32), {}

    def step(self, action):
        """Takes a single step in the episode given `action`.

        Returns:
            New observation, reward, terminated-flag, truncated-flag, info-dict (empty).
        """
        # Walk left.
        if action == 0 and self.cur_pos > 0:
            self.cur_pos -= 1
        # Walk right.
        elif action == 1:
            self.cur_pos += 1
        # Set `terminated` flag when end of corridor (goal) reached.
        terminated = self.cur_pos >= self.end_pos
        truncated = False
        # +1 when goal reached, otherwise -1.
        reward = 1.0 if terminated else -0.1
        return np.array([self.cur_pos], np.float32), reward, terminated, truncated, {}


# Create an RLlib Algorithm instance from a PPOConfig object.
config = (
    PPOConfig().environment(
        # Env class to use (here: our gym.Env sub-class from above).
        SimpleCorridor,
        # Config dict to be passed to our custom env's constructor.
        # Use corridor with 20 fields (including S and G).
        env_config={"corridor_length": 20},
    )
    # Parallelize environment rollouts.
    .env_runners(num_env_runners=3)
)
# Construct the actual (PPO) algorithm object from the config.
algo = config.build()
rl_module = algo.get_module()

# Train for n iterations and report results (mean episode rewards).
# Since we have to move at least 19 times in the env to reach the goal and
# each move gives us -0.1 reward (except the last move at the end: +1.0),
# Expect to reach an optimal episode reward of `-0.1*18 + 1.0 = -0.8`.
for i in range(5):
    results = algo.train()
    print(f"Iter: {i}; avg. results={results['env_runners']}")

# Perform inference (action computations) based on given env observations.
# Note that we are using a slightly different env here (len 10 instead of 20),
# however, this should still work as the agent has (hopefully) learned
# to "just always walk right!"
env = SimpleCorridor({"corridor_length": 10})
# Get the initial observation (should be: [0.0] for the starting position).
obs, info = env.reset()
terminated = truncated = False
total_reward = 0.0
# Play one episode.
while not terminated and not truncated:
    # Compute a single action, given the current observation
    # from the environment.
    action_logits = rl_module.forward_inference(
        {"obs": torch.from_numpy(obs).unsqueeze(0)}
    )["action_dist_inputs"].numpy()[
        0
    ]  # [0]: B=1
    action = np.argmax(action_logits)
    # Apply the computed action in the environment.
    obs, reward, terminated, truncated, info = env.step(action)
    # Sum up rewards for reporting purposes.
    total_reward += reward
# Report results.
print(f"Played 1 episode; total-reward={total_reward}")