Distributed LightGBM on Ray¶

LightGBM-Ray is a distributed backend for LightGBM, built on top of distributed computing framework Ray.

LightGBM-Ray

enables multi-node and multi-GPU training
integrates seamlessly with distributed hyperparameter optimization library Ray Tune
comes with fault tolerance handling mechanisms, and
supports distributed dataframes and distributed data loading

All releases are tested on large clusters and workloads.

This package is based on XGBoost-Ray. As of now, XGBoost-Ray is a dependency for LightGBM-Ray.

Installation¶

You can install the latest LightGBM-Ray release from PIP:

pip install lightgbm_ray

If you’d like to install the latest master, use this command instead:

pip install git+https://github.com/ray-project/lightgbm_ray.git#lightgbm_ray

Usage¶

LightGBM-Ray provides a drop-in replacement for LightGBM’s train function. To pass data, a RayDMatrix object is required, common with XGBoost-Ray. You can also use a scikit-learn interface - see next section.

Just as in original lgbm.train() function, the training parameters are passed as the params dictionary.

Ray-specific distributed training parameters are configured with a lightgbm_ray.RayParams object. For instance, you can set the num_actors property to specify how many distributed actors you would like to use.

Here is a simplified example (which requires sklearn):

Training:

from lightgbm_ray import RayDMatrix, RayParams, train
from sklearn.datasets import load_breast_cancer

train_x, train_y = load_breast_cancer(return_X_y=True)
train_set = RayDMatrix(train_x, train_y)

evals_result = {}
bst = train(
    {
        "objective": "binary",
        "metric": ["binary_logloss", "binary_error"],
    },
    train_set,
    evals_result=evals_result,
    valid_sets=[train_set],
    valid_names=["train"],
    verbose_eval=False,
    ray_params=RayParams(num_actors=2, cpus_per_actor=2))

bst.booster_.save_model("model.lgbm")
print("Final training error: {:.4f}".format(
    evals_result["train"]["binary_error"][-1]))

Prediction:

from lightgbm_ray import RayDMatrix, RayParams, predict
from sklearn.datasets import load_breast_cancer
import lightgbm as lgbm

data, labels = load_breast_cancer(return_X_y=True)

dpred = RayDMatrix(data, labels)

bst = lgbm.Booster(model_file="model.lgbm")
pred_ray = predict(bst, dpred, ray_params=RayParams(num_actors=2))

print(pred_ray)

scikit-learn API¶

LightGBM-Ray also features a scikit-learn API fully mirroring pure LightGBM scikit-learn API, providing a completely drop-in replacement. The following estimators are available:

RayLGBMClassifier
RayLGBMRegressor

Example usage of RayLGBMClassifier:

from lightgbm_ray import RayLGBMClassifier, RayParams
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split

seed = 42

X, y = load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(
    X, y, train_size=0.25, random_state=42)

clf = RayLGBMClassifier(
    n_jobs=2,  # In LightGBM-Ray, n_jobs sets the number of actors
    random_state=seed)

# scikit-learn API will automatically convert the data
# to RayDMatrix format as needed.
# You can also pass X as a RayDMatrix, in which case
# y will be ignored.

clf.fit(X_train, y_train)

pred_ray = clf.predict(X_test)
print(pred_ray)

pred_proba_ray = clf.predict_proba(X_test)
print(pred_proba_ray)

# It is also possible to pass a RayParams object
# to fit/predict/predict_proba methods - will override
# n_jobs set during initialization

clf.fit(X_train, y_train, ray_params=RayParams(num_actors=2))

pred_ray = clf.predict(X_test, ray_params=RayParams(num_actors=2))
print(pred_ray)

Things to keep in mind:

n_jobs parameter controls the number of actors spawned. You can pass a RayParams object to the fit/predict/predict_proba methods as the ray_params argument for greater control over resource allocation. Doing so will override the value of n_jobs with the value of ray_params.num_actors attribute. For more information, refer to the Resources section below.
By default n_jobs is set to 1, which means the training will not be distributed. Make sure to either set n_jobs to a higher value or pass a RayParams object as outlined above in order to take advantage of LightGBM-Ray’s functionality.
After calling fit, additional evaluation results (e.g. training time, number of rows, callback results) will be available under additional_results_ attribute.
eval_ arguments are supported, but early stopping is not.
LightGBM-Ray’s scikit-learn API is based on LightGBM 3.2.1. While we try to support older LightGBM versions, please note that this library is only fully tested and supported for LightGBM >= 3.2.1.

For more information on the scikit-learn API, refer to the LightGBM documentation.

Data loading¶

Data is passed to LightGBM-Ray via a RayDMatrix object.

The RayDMatrix lazy loads data and stores it sharded in the Ray object store. The Ray LightGBM actors then access these shards to run their training on.

A RayDMatrix support various data and file types, like Pandas DataFrames, Numpy Arrays, CSV files and Parquet files.

Example loading multiple parquet files:

import glob
from lightgbm_ray import RayDMatrix, RayFileType

# We can also pass a list of files
path = list(sorted(glob.glob("/data/nyc-taxi/*/*/*.parquet")))

# This argument will be passed to `pd.read_parquet()`
columns = [
    "passenger_count",
    "trip_distance", "pickup_longitude", "pickup_latitude",
    "dropoff_longitude", "dropoff_latitude",
    "fare_amount", "extra", "mta_tax", "tip_amount",
    "tolls_amount", "total_amount"
]

dtrain = RayDMatrix(
    path,
    label="passenger_count",  # Will select this column as the label
    columns=columns,
    # ignore=["total_amount"],  # Optional list of columns to ignore
    filetype=RayFileType.PARQUET)

Hyperparameter Tuning¶

LightGBM-Ray integrates with Ray Tune to provide distributed hyperparameter tuning for your distributed LightGBM models. You can run multiple LightGBM-Ray training runs in parallel, each with a different hyperparameter configuration, and each training run parallelized by itself. All you have to do is move your training code to a function, and pass the function to tune.run. Internally, train will detect if tune is being used and will automatically report results to tune.

Example using LightGBM-Ray with Ray Tune:

from lightgbm_ray import RayDMatrix, RayParams, train
from sklearn.datasets import load_breast_cancer

num_actors = 2
num_cpus_per_actor = 2

ray_params = RayParams(
    num_actors=num_actors, cpus_per_actor=num_cpus_per_actor)

def train_model(config):
    train_x, train_y = load_breast_cancer(return_X_y=True)
    train_set = RayDMatrix(train_x, train_y)

    evals_result = {}
    bst = train(
        params=config,
        dtrain=train_set,
        evals_result=evals_result,
        valid_sets=[train_set],
        valid_names=["train"],
        verbose_eval=False,
        ray_params=ray_params)
    bst.booster_.save_model("model.lgbm")

from ray import tune

# Specify the hyperparameter search space.
config = {
    "objective": "binary",
    "metric": ["binary_logloss", "binary_error"],
    "eta": tune.loguniform(1e-4, 1e-1),
    "subsample": tune.uniform(0.5, 1.0),
    "max_depth": tune.randint(1, 9)
}

# Make sure to use the `get_tune_resources` method to set the `resources_per_trial`
analysis = tune.run(
    train_model,
    config=config,
    metric="train-binary_error",
    mode="min",
    num_samples=4,
    resources_per_trial=ray_params.get_tune_resources())
print("Best hyperparameters", analysis.best_config)

Also see examples/simple_tune.py for another example.

Fault tolerance¶

LightGBM-Ray leverages the stateful Ray actor model to enable fault tolerant training. Currently, only non-elastic training is supported.

Non-elastic training (warm restart)¶

When an actor or node dies, LightGBM-Ray will retain the state of the remaining actors. In non-elastic training, the failed actors will be replaced as soon as resources are available again. Only these actors will reload their parts of the data. Training will resume once all actors are ready for training again.

You can configure this mode in the RayParams:

from lightgbm_ray import RayParams

ray_params = RayParams(
    max_actor_restarts=2,    # How often are actors allowed to fail, Default = 0
)

Resources¶

By default, LightGBM-Ray tries to determine the number of CPUs available and distributes them evenly across actors.

It is important to note that distributed LightGBM needs at least two CPUs per actor to function efficiently (without blocking). Therefore, by default, at least two CPUs will be assigned to each actor, and an exception will be raised if an actor has less than two CPUs. It is possible to override this check by setting the allow_less_than_two_cpus argument to True, though it is not recommended, as it will negatively impact training performance.

In the case of very large clusters or clusters with many different machine sizes, it makes sense to limit the number of CPUs per actor by setting the cpus_per_actor argument. Consider always setting this explicitly.

The number of LightGBM actors always has to be set manually with the num_actors argument.

Multi GPU training¶

LightGBM-Ray enables multi GPU training. The LightGBM core backend will automatically handle communication. All you have to do is to start one actor per GPU and set LightGBM’s device_type to a GPU-compatible option, eg. gpu (see LightGBM documentation for more details.)

For instance, if you have 2 machines with 4 GPUs each, you will want to start 8 remote actors, and set gpus_per_actor=1. There is usually no benefit in allocating less (e.g. 0.5) or more than one GPU per actor.

You should divide the CPUs evenly across actors per machine, so if your machines have 16 CPUs in addition to the 4 GPUs, each actor should have 4 CPUs to use.

from lightgbm_ray import RayParams

ray_params = RayParams(
    num_actors=8,
    gpus_per_actor=1,
    cpus_per_actor=4,   # Divide evenly across actors per machine
)

How many remote actors should I use?¶

This depends on your workload and your cluster setup. Generally there is no inherent benefit of running more than one remote actor per node for CPU-only training. This is because LightGBM core can already leverage multiple CPUs via threading.

However, there are some cases when you should consider starting more than one actor per node:

For multi GPU training, each GPU should have a separate remote actor. Thus, if your machine has 24 CPUs and 4 GPUs, you will want to start 4 remote actors with 6 CPUs and 1 GPU each
In a heterogeneous cluster, you might want to find the greatest common divisor for the number of CPUs. E.g. for a cluster with three nodes of 4, 8, and 12 CPUs, respectively, you should set the number of actors to 6 and the CPUs per actor to 4.

Distributed data loading¶

LightGBM-Ray can leverage both centralized and distributed data loading.

In centralized data loading, the data is partitioned by the head node and stored in the object store. Each remote actor then retrieves their partitions by querying the Ray object store. Centralized loading is used when you pass centralized in-memory dataframes, such as Pandas dataframes or Numpy arrays, or when you pass a single source file, such as a single CSV or Parquet file.

from lightgbm_ray import RayDMatrix

# This will use centralized data loading, as only one source file is specified
# `label_col` is a column in the CSV, used as the target label
ray_params = RayDMatrix("./source_file.csv", label="label_col")

In distributed data loading, each remote actor loads their data directly from the source (e.g. local hard disk, NFS, HDFS, S3), without a central bottleneck. The data is still stored in the object store, but locally to each actor. This mode is used automatically when loading data from multiple CSV or Parquet files. Please note that we do not check or enforce partition sizes in this case - it is your job to make sure the data is evenly distributed across the source files.

from lightgbm_ray import RayDMatrix

# This will use distributed data loading, as four source files are specified
# Please note that you cannot schedule more than four actors in this case.
# `label_col` is a column in the Parquet files, used as the target label
ray_params = RayDMatrix([
    "hdfs:///tmp/part1.parquet",
    "hdfs:///tmp/part2.parquet",
    "hdfs:///tmp/part3.parquet",
    "hdfs:///tmp/part4.parquet",
], label="label_col")

Lastly, LightGBM-Ray supports distributed dataframe representations, such as Ray Datasets, Modin and Dask dataframes (used with Dask on Ray). Here, LightGBM-Ray will check on which nodes the distributed partitions are currently located, and will assign partitions to actors in order to minimize cross-node data transfer. Please note that we also assume here that partition sizes are uniform.

from lightgbm_ray import RayDMatrix

# This will try to allocate the existing Modin partitions
# to co-located Ray actors. If this is not possible, data will
# be transferred across nodes
ray_params = RayDMatrix(existing_modin_df)

Data sources¶

The following data sources can be used with a RayDMatrix object.

Type	Centralized loading	Distributed loading
Numpy array	Yes	No
Pandas dataframe	Yes	No
Single CSV	Yes	No
Multi CSV	Yes	Yes
Single Parquet	Yes	No
Multi Parquet	Yes	Yes
Ray Dataset	Yes	Yes
Petastorm	Yes	Yes
Dask dataframe	Yes	Yes
Modin dataframe	Yes	Yes

Memory usage¶

Details coming soon.

Best practices

In order to reduce peak memory usage, consider the following suggestions:

Store data as float32 or less. More precision is often not needed, and keeping data in a smaller format will help reduce peak memory usage for initial data loading.
Pass the dtype when loading data from CSV. Otherwise, floating point values will be loaded as np.float64 per default, increasing peak memory usage by 33%.

Placement Strategies¶

LightGBM-Ray leverages Ray’s Placement Group API (https://docs.ray.io/en/master/placement-group.html) to implement placement strategies for better fault tolerance.

By default, a SPREAD strategy is used for training, which attempts to spread all of the training workers across the nodes in a cluster on a best-effort basis. This improves fault tolerance since it minimizes the number of worker failures when a node goes down, but comes at a cost of increased inter-node communication To disable this strategy, set the RXGB_USE_SPREAD_STRATEGY environment variable to 0. If disabled, no particular placement strategy will be used.

When LightGBM-Ray is used with Ray Tune for hyperparameter tuning, a PACK strategy is used. This strategy attempts to place all workers for each trial on the same node on a best-effort basis. This means that if a node goes down, it will be less likely to impact multiple trials.

When placement strategies are used, LightGBM-Ray will wait for 100 seconds for the required resources to become available, and will fail if the required resources cannot be reserved and the cluster cannot autoscale to increase the number of resources. You can change the RXGB_PLACEMENT_GROUP_TIMEOUT_S environment variable to modify how long this timeout should be.

More examples¶

For complete end to end examples, please have a look at the examples folder:

Simple sklearn breastcancer dataset example (requires sklearn)
HIGGS classification example (download dataset (2.6 GB))
HIGGS classification example with Parquet (uses the same dataset)
Test data classification (uses a self-generated dataset)

API reference¶

class lightgbm_ray.RayParams(num_actors: int = 0, cpus_per_actor: int = 0, gpus_per_actor: int = - 1, resources_per_actor: Optional[Dict] = None, elastic_training: bool = False, max_failed_actors: int = 0, max_actor_restarts: int = 0, checkpoint_frequency: int = 5, distributed_callbacks: Optional[List[xgboost_ray.callback.DistributedCallback]] = None, allow_less_than_two_cpus: bool = False)[source]¶

Parameters to configure Ray-specific behavior.

Parameters

num_actors (int) – Number of parallel Ray actors.
cpus_per_actor (int) – Number of CPUs to be used per Ray actor. If smaller than 2, training might be substantially slower because communication work and training work will block each other. This will raise an exception unless allow_less_than_two_cpus is True.
gpus_per_actor (int) – Number of GPUs to be used per Ray actor.
resources_per_actor (Optional[Dict]) – Dict of additional resources required per Ray actor.
allow_less_than_two_cpus (bool) – If True, an exception will not be raised if cpus_per_actor. Default False.
max_failed_actors (int) – If elastic_training is True, this specifies the maximum number of failed actors with which we still continue training.
max_actor_restarts (int) – Number of retries when Ray actors fail. Defaults to 0 (no retries). Set to -1 for unlimited retries.
checkpoint_frequency (int) – How often to save checkpoints. Defaults to 5 (every 5th iteration).

PublicAPI (beta): This API is in beta and may change before becoming stable.

get_tune_resources()[source]¶: Return the resources to use for xgboost_ray training with Tune.

Note

The xgboost_ray.RayDMatrix class is shared with XGBoost-Ray.

class xgboost_ray.RayDMatrix(data: Union[str, List[str], numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series, ray.util.data.dataset.MLDataset], label: Optional[Union[str, List[str], numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series, ray.util.data.dataset.MLDataset]] = None, weight: Optional[Union[str, List[str], numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series, ray.util.data.dataset.MLDataset]] = None, base_margin: Optional[Union[str, List[str], numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series, ray.util.data.dataset.MLDataset]] = None, missing: Optional[float] = None, label_lower_bound: Optional[Union[str, List[str], numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series, ray.util.data.dataset.MLDataset]] = None, label_upper_bound: Optional[Union[str, List[str], numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series, ray.util.data.dataset.MLDataset]] = None, feature_names: Optional[List[str]] = None, feature_types: Optional[List[numpy.dtype]] = None, qid: Optional[Union[str, List[str], numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series, ray.util.data.dataset.MLDataset]] = None, num_actors: Optional[int] = None, filetype: Optional[xgboost_ray.data_sources.data_source.RayFileType] = None, ignore: Optional[List[str]] = None, distributed: Optional[bool] = None, sharding: xgboost_ray.matrix.RayShardingMode = RayShardingMode.INTERLEAVED, lazy: bool = False, **kwargs)[source]

XGBoost on Ray DMatrix class.

This is the data object that the training and prediction functions expect. This wrapper manages distributed data by sharding the data for the workers and storing the shards in the object store.

If this class is called without the num_actors argument, it will be lazy loaded. Thus, it will return immediately and only load the data and store it in the Ray object store after load_data(num_actors) or get_data(rank, num_actors) is called.

If this class is instantiated with the num_actors argument, it will directly load the data and store them in the object store. If this should be deferred, pass lazy=True as an argument.

Loading the data will store it in the Ray object store. This object then stores references to the data shards in the Ray object store. Actors can request these shards with the get_data(rank) method, returning dataframes according to the actor rank.

The total number of actors has to remain constant and cannot be changed once it has been set.

Parameters

data – Data object. Can be a pandas dataframe, pandas series, numpy array, Ray MLDataset, modin dataframe, string pointing to a csv or parquet file, or list of strings pointing to csv or parquet files.
label – Optional label object. Can be a pandas series, numpy array, modin series, string pointing to a csv or parquet file, or a string indicating the column of the data dataframe that contains the label. If this is not a string it must be of the same type as the data argument.
num_actors – Number of actors to shard this data for. If this is not None, data will be loaded and stored into the object store after initialization. If this is None, it will be set by the xgboost_ray.train() function, and it will be loaded and stored in the object store then. Defaults to None (
filetype (Optional[RayFileType]) – Type of data to read. This is disregarded if a data object like a pandas dataframe is passed as the data argument. For filenames, the filetype is automaticlly detected via the file name (e.g. .csv will be detected as RayFileType.CSV). Passing this argument will overwrite the detected filename. If the filename cannot be determined from the data object, passing this is mandatory. Defaults to None (auto detection).
ignore (Optional[List[str]]) – Exclude these columns from the dataframe after loading the data.
distributed (Optional[bool]) – If True, use distributed loading (each worker loads a share of the dataset). If False, use central loading (the head node loads the whole dataset and distributed it). If None, auto-detect and default to distributed loading, if possible.
sharding (RayShardingMode) – How to shard the data for different workers. RayShardingMode.INTERLEAVED will divide the data per row, i.e. every i-th row will be passed to the first worker, every (i+1)th row to the second worker, etc. RayShardingMode.BATCH will divide the data in batches, i.e. the first 0-(m-1) rows will be passed to the first worker, the m-(2m-1) rows to the second worker, etc. Defaults to RayShardingMode.INTERLEAVED. If using distributed data loading, sharding happens on a per-file basis, and not on a per-row basis, i.e. For interleaved every ith file will be passed into the first worker, etc.
lazy (bool) – If num_actors is passed, setting this to True will defer data loading and storing until load_data() or get_data() is called. Defaults to False.
**kwargs – Keyword arguments will be passed to the data loading function. For instance, with RayFileType.PARQUET, these arguments will be passed to pandas.read_parquet().

from xgboost_ray import RayDMatrix, RayFileType

files = ["data_one.parquet", "data_two.parquet"]

columns = ["feature_1", "feature_2", "label_column"]

dtrain = RayDMatrix(
    files,
    num_actors=4,  # Will shard the data for four workers
    label="label_column",  # Will select this column as the label
    columns=columns,  # Will be passed to `pandas.read_parquet()`
    filetype=RayFileType.PARQUET)

PublicAPI (beta): This API is in beta and may change before becoming stable.

load_data(num_actors: Optional[int] = None, rank: Optional[int] = None)[source]

Load data, putting it into the Ray object store.

If a rank is given, only data for this rank is loaded (for distributed data sources only).

get_data(rank: int, num_actors: Optional[int] = None) → Dict[str, Union[None, pandas.core.frame.DataFrame, List[Optional[pandas.core.frame.DataFrame]]]][source]

Get data, i.e. return dataframe for a specific actor.

This method is called from an actor, given its rank and the total number of actors. If the data is not yet loaded, loading is triggered.

unload_data()[source]: Delete object references to clear object store

lightgbm_ray.train(params: Dict, dtrain: xgboost_ray.matrix.RayDMatrix, model_factory: Type[lightgbm.sklearn.LGBMModel] = <class 'lightgbm.sklearn.LGBMModel'>, num_boost_round: int = 10, *args, valid_sets: Optional[List[xgboost_ray.matrix.RayDMatrix]] = None, valid_names: Optional[List[str]] = None, verbose_eval: Union[bool, int] = True, evals: Union[List[Tuple[xgboost_ray.matrix.RayDMatrix, str]], Tuple[xgboost_ray.matrix.RayDMatrix, str]] = (), evals_result: Optional[Dict] = None, additional_results: Optional[Dict] = None, ray_params: Union[None, lightgbm_ray.main.RayParams, Dict] = None, _remote: Optional[bool] = None, **kwargs) → lightgbm.sklearn.LGBMModel[source]¶

Distributed LightGBM training via Ray.

This function will connect to a Ray cluster, create num_actors remote actors, send data shards to them, and have them train an LightGBM model using LightGBM’s built-in distributed mode.

This method handles setting up the following network parameters: - local_listen_port: port that each LightGBM worker opens a listening socket on, to accept connections from other workers. This can differ from LightGBM worker to LightGBM worker, but does not have to. - machines: a comma-delimited list of all workers in the cluster, in the form ip:port,ip:port. If running multiple workers on the same Ray Node, use different ports for each worker. For example, for ray_params.num_actors=3, you might pass "127.0.0.1:12400,127.0.0.1:12401,127.0.0.1:12402".

The default behavior of this function is to generate machines based on Ray workers, and to search for an open port on each worker to be used as local_listen_port.

If machines is provided explicitly in params, this function uses the hosts and ports in that list directly, and will try to start Ray workers on the nodes with the given ips. If that is not possible, or any of those ports are not free when training starts, training will fail.

If local_listen_port is provided in params and machines is not, this function constructs machines automatically from auto-assigned Ray workers, assuming that each one will use the same local_listen_port.

Failure handling:

LightGBM on Ray supports automatic failure handling that can be configured with the ray_params argument. If an actor or local training task dies, the Ray actor is marked as dead and the number of restarts is below ray_params.max_actor_restarts, Ray will try to schedule the dead actor again, load the data shard on this actor, and then continue training from the latest checkpoint.

Otherwise, training is aborted.

Parameters

params (Dict) – parameter dict passed to LGBMModel
dtrain (RayDMatrix) – Data object containing the training data.
model_factory (Type[LGBMModel]) –
valid_sets (Optional[List[RayDMatrix]]) – List of data to be evaluated on during training. Mutually exclusive with evals.
Optional[List[str]] (valid_names) – Names of valid_sets.
evals (Union[List[Tuple[RayDMatrix, str]], Tuple[RayDMatrix, str]]) – evals tuple passed to LGBMModel.fit(). Mutually exclusive with valid_sets.
evals_result (Optional[Dict]) – Dict to store evaluation results in.
verbose_eval (Union[bool, int]) – Requires at least one validation data. If True, the eval metric on the valid set is printed at each boosting stage. If int, the eval metric on the valid set is printed at every verbose_eval boosting stage. The last boosting stage or the boosting stage found by using early_stopping_rounds is also printed. With verbose_eval = 4 and at least one item in valid_sets, an evaluation metric is printed every 4 (instead of 1) boosting stages.
additional_results (Optional[Dict]) – Dict to store additional results.
ray_params (Union[None, lightgbm_ray.RayParams, Dict]) – Parameters to configure Ray-specific behavior. See lightgbm_ray.RayParams for a list of valid configuration parameters.
_remote (bool) – Whether to run the driver process in a remote function. This is enabled by default in Ray client mode.
**kwargs – Keyword arguments will be passed to the local model_factory.fit() calls.

Returns: An LGBMModel object.

PublicAPI (beta): This API is in beta and may change before becoming stable.

lightgbm_ray.predict(model: Union[lightgbm.sklearn.LGBMModel, lightgbm.basic.Booster], data: xgboost_ray.matrix.RayDMatrix, method: str = 'predict', ray_params: Union[None, lightgbm_ray.main.RayParams, Dict] = None, _remote: Optional[bool] = None, **kwargs) → Optional[numpy.ndarray][source]¶

Distributed LightGBM predict via Ray.

This function will connect to a Ray cluster, create num_actors remote actors, send data shards to them, and have them predict labels using an LightGBM model. The results are then combined and returned.

Parameters

model (Union[LGBMModel, Booster]) – Model or booster object to call for prediction.
data (RayDMatrix) – Data object containing the prediction data.
method (str) – Name of estimator method to use for prediction.
ray_params (Union[None, lightgbm_ray.RayParams, Dict]) – Parameters to configure Ray-specific behavior. See lightgbm_ray.RayParams for a list of valid configuration parameters.
_remote (bool) – Whether to run the driver process in a remote function. This is enabled by default in Ray client mode.
**kwargs – Keyword arguments will be passed to the local xgb.predict() calls.

Returns: np.ndarray containing the predicted labels.

PublicAPI (beta): This API is in beta and may change before becoming stable.

scikit-learn API¶

class lightgbm_ray.RayLGBMClassifier(boosting_type: str = 'gbdt', num_leaves: int = 31, max_depth: int = - 1, learning_rate: float = 0.1, n_estimators: int = 100, subsample_for_bin: int = 200000, objective: Optional[Union[str, Callable]] = None, class_weight: Optional[Union[Dict, str]] = None, min_split_gain: float = 0.0, min_child_weight: float = 0.001, min_child_samples: int = 20, subsample: float = 1.0, subsample_freq: int = 0, colsample_bytree: float = 1.0, reg_alpha: float = 0.0, reg_lambda: float = 0.0, random_state: Optional[Union[int, numpy.random.mtrand.RandomState]] = None, n_jobs: int = - 1, silent: Union[bool, str] = 'warn', importance_type: str = 'split', **kwargs)[source]¶

PublicAPI (beta): This API is in beta and may change before becoming stable.

fit(X, y, sample_weight=None, init_score=None, eval_set=None, eval_names: Optional[List[str]] = None, eval_sample_weight=None, eval_class_weight: Optional[List[Union[dict, str]]] = None, eval_init_score=None, eval_metric: Optional[Union[Callable, str, List[Union[Callable, str]]]] = None, ray_params: Union[None, lightgbm_ray.main.RayParams, Dict] = None, _remote: Optional[bool] = None, ray_dmatrix_params: Optional[Dict] = None, **kwargs: Any) → lightgbm_ray.sklearn.RayLGBMClassifier[source]¶

Build a gradient boosting model from the training set (X, y).

Parameters

X (array-like or sparse matrix of shape = [n_samples, n_features]) – Input feature matrix.
y (array-like of shape = [n_samples]) – The target values (class labels in classification, real numbers in regression).
sample_weight (array-like of shape = [n_samples] or None, optional (default=None)) – Weights of training data.
init_score (array-like of shape = [n_samples] or shape = [n_samples * n_classes] (for multi-class task) or shape = [n_samples, n_classes] (for multi-class task) or None, optional (default=None)) – Init score of training data.
eval_set (list or None, optional (default=None)) – A list of (X, y) tuple pairs to use as validation sets.
eval_names (list of str, or None, optional (default=None)) – Names of eval_set.
eval_sample_weight (list of array, or None, optional (default=None)) – Weights of eval data.
eval_class_weight (list or None, optional (default=None)) – Class weights of eval data.
eval_init_score (list of array, or None, optional (default=None)) – Init score of eval data.
eval_metric (str, callable, list or None, optional (default=None)) – If str, it should be a built-in evaluation metric to use. If callable, it should be a custom evaluation metric, see note below for more details. If list, it can be a list of built-in metrics, a list of custom evaluation metrics, or a mix of both. In either case, the metric from the model parameters will be evaluated and used as well. Default: ‘l2’ for LGBMRegressor, ‘logloss’ for LGBMClassifier, ‘ndcg’ for LGBMRanker.
early_stopping_rounds (int or None, optional (default=None)) – Activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training. Requires at least one validation data and one metric. If there’s more than one, will check all of them. But the training data is ignored anyway. To check only the first metric, set the first_metric_only parameter to True in additional parameters **kwargs of the model constructor.
verbose (bool or int, optional (default=True)) –
Requires at least one evaluation data. If True, the eval metric on the eval set is printed at each boosting stage. If int, the eval metric on the eval set is printed at every verbose boosting stage. The last boosting stage or the boosting stage found by using early_stopping_rounds is also printed.

Example

With verbose = 4 and at least one item in eval_set, an evaluation metric is printed every 4 (instead of 1) boosting stages.
feature_name (list of str, or 'auto', optional (default='auto')) – Feature names. If ‘auto’ and data is pandas DataFrame, data columns names are used.
categorical_feature (list of str or int, or 'auto', optional (default='auto')) – Categorical features. If list of int, interpreted as indices. If list of str, interpreted as feature names (need to specify feature_name as well). If ‘auto’ and data is pandas DataFrame, pandas unordered categorical columns are used. All values in categorical features should be less than int32 max value (2147483647). Large values could be memory consuming. Consider using consecutive integers starting from zero. All negative values in categorical features will be treated as missing values. The output cannot be monotonically constrained with respect to a categorical feature.
callbacks (list of callable, or None, optional (default=None)) – List of callback functions that are applied at each iteration. See Callbacks in Python API for more information.
init_model (str, pathlib.Path, Booster, LGBMModel or None, optional (default=None)) – Filename of LightGBM model, Booster instance or LGBMModel instance used for continue training.
ray_params (lightgbm_ray.RayParams or dict, optional (default=None)) – Parameters to configure Ray-specific behavior. See lightgbm_ray.RayParams for a list of valid configuration parameters. Will override n_jobs attribute with own num_actors parameter.
_remote (bool, optional (default=False)) – Whether to run the driver process in a remote function. This is enabled by default in Ray client mode.
ray_dmatrix_params (dict, optional (default=None)) – Dict of parameters (such as sharding mode) passed to the internal RayDMatrix initialization.

Returns

self – Returns self.

Return type

object

Note

Custom eval function expects a callable with following signatures: func(y_true, y_pred), func(y_true, y_pred, weight) or func(y_true, y_pred, weight, group) and returns (eval_name, eval_result, is_higher_better) or list of (eval_name, eval_result, is_higher_better):

y_truearray-like of shape = [n_samples]
The target values.

y_predarray-like of shape = [n_samples] or shape = [n_samples * n_classes] (for multi-class task)
The predicted values. In case of custom objective, predicted values are returned before any transformation, e.g. they are raw margin instead of probability of positive class for binary task in this case.

weightarray-like of shape = [n_samples]
The weight of samples.

grouparray-like
Group/query data. Only used in the learning-to-rank task. sum(group) = n_samples. For example, if you have a 100-document dataset with group = [10, 20, 40, 10, 10, 10], that means that you have 6 groups, where the first 10 records are in the first group, records 11-30 are in the second group, records 31-70 are in the third group, etc.

eval_namestr
The name of evaluation function (without whitespace).

eval_resultfloat
The eval result.

is_higher_betterbool
Is eval result higher better, e.g. AUC is is_higher_better.

For multi-class task, the y_pred is group by class_id first, then group by row_id. If you want to get i-th row y_pred in j-th class, the access way is y_pred[j * num_data + i].

predict_proba(X, *, ray_params: Union[None, lightgbm_ray.main.RayParams, Dict] = None, _remote: Optional[bool] = None, ray_dmatrix_params: Optional[Dict] = None, **kwargs)[source]¶

Return the predicted probability for each class for each sample.

Parameters

X (array-like or sparse matrix of shape = [n_samples, n_features]) – Input features matrix.
raw_score (bool, optional (default=False)) – Whether to predict raw scores.
start_iteration (int, optional (default=0)) – Start index of the iteration to predict. If <= 0, starts from the first iteration.
num_iteration (int or None, optional (default=None)) – Total number of iterations used in the prediction. If None, if the best iteration exists and start_iteration <= 0, the best iteration is used; otherwise, all iterations from start_iteration are used (no limits). If <= 0, all iterations from start_iteration are used (no limits).
pred_leaf (bool, optional (default=False)) – Whether to predict leaf index.
pred_contrib (bool, optional (default=False)) –
Whether to predict feature contributions.

Note

If you want to get more explanations for your model’s predictions using SHAP values, like SHAP interaction values, you can install the shap package (https://github.com/slundberg/shap). Note that unlike the shap package, with pred_contrib we return a matrix with an extra column, where the last column is the expected value.
ray_params (lightgbm_ray.RayParams or dict, optional (default=None)) – Parameters to configure Ray-specific behavior. See lightgbm_ray.RayParams for a list of valid configuration parameters. Will override n_jobs attribute with own num_actors parameter.
_remote (bool, optional (default=False)) – Whether to run the driver process in a remote function. This is enabled by default in Ray client mode.
ray_dmatrix_params (dict, optional (default=None)) – Dict of parameters (such as sharding mode) passed to the internal RayDMatrix initialization.
**kwargs – Other parameters for the prediction.

Returns

predicted_probability (array-like of shape = [n_samples] or shape = [n_samples, n_classes]) – The predicted values.
X_leaves (array-like of shape = [n_samples, n_trees] or shape = [n_samples, n_trees * n_classes]) – If pred_leaf=True, the predicted leaf of every tree for each sample.
X_SHAP_values (array-like of shape = [n_samples, n_features + 1] or shape = [n_samples, (n_features + 1) * n_classes] or list with n_classes length of such objects) – If pred_contrib=True, the feature contributions for each sample.

predict(X, *, ray_params: Union[None, lightgbm_ray.main.RayParams, Dict] = None, _remote: Optional[bool] = None, ray_dmatrix_params: Optional[Dict] = None, **kwargs)[source]¶

Return the predicted value for each sample.

Parameters

X (array-like or sparse matrix of shape = [n_samples, n_features]) – Input features matrix.
raw_score (bool, optional (default=False)) – Whether to predict raw scores.
start_iteration (int, optional (default=0)) – Start index of the iteration to predict. If <= 0, starts from the first iteration.
num_iteration (int or None, optional (default=None)) – Total number of iterations used in the prediction. If None, if the best iteration exists and start_iteration <= 0, the best iteration is used; otherwise, all iterations from start_iteration are used (no limits). If <= 0, all iterations from start_iteration are used (no limits).
pred_leaf (bool, optional (default=False)) – Whether to predict leaf index.
pred_contrib (bool, optional (default=False)) –
Whether to predict feature contributions.

Note

If you want to get more explanations for your model’s predictions using SHAP values, like SHAP interaction values, you can install the shap package (https://github.com/slundberg/shap). Note that unlike the shap package, with pred_contrib we return a matrix with an extra column, where the last column is the expected value.
ray_params (lightgbm_ray.RayParams or dict, optional (default=None)) – Parameters to configure Ray-specific behavior. See lightgbm_ray.RayParams for a list of valid configuration parameters. Will override n_jobs attribute with own num_actors parameter.
_remote (bool, optional (default=False)) – Whether to run the driver process in a remote function. This is enabled by default in Ray client mode.
ray_dmatrix_params (dict, optional (default=None)) – Dict of parameters (such as sharding mode) passed to the internal RayDMatrix initialization.
**kwargs – Other parameters for the prediction.

Returns

predicted_result (array-like of shape = [n_samples] or shape = [n_samples, n_classes]) – The predicted values.
X_leaves (array-like of shape = [n_samples, n_trees] or shape = [n_samples, n_trees * n_classes]) – If pred_leaf=True, the predicted leaf of every tree for each sample.
X_SHAP_values (array-like of shape = [n_samples, n_features + 1] or shape = [n_samples, (n_features + 1) * n_classes] or list with n_classes length of such objects) – If pred_contrib=True, the feature contributions for each sample.

to_local() → lightgbm.sklearn.LGBMClassifier[source]¶

Create regular version of lightgbm.LGBMClassifier from the distributed version.

Returns: model – Local underlying model.
Return type: lightgbm.LGBMClassifier

class lightgbm_ray.RayLGBMRegressor(boosting_type: str = 'gbdt', num_leaves: int = 31, max_depth: int = - 1, learning_rate: float = 0.1, n_estimators: int = 100, subsample_for_bin: int = 200000, objective: Optional[Union[str, Callable]] = None, class_weight: Optional[Union[Dict, str]] = None, min_split_gain: float = 0.0, min_child_weight: float = 0.001, min_child_samples: int = 20, subsample: float = 1.0, subsample_freq: int = 0, colsample_bytree: float = 1.0, reg_alpha: float = 0.0, reg_lambda: float = 0.0, random_state: Optional[Union[int, numpy.random.mtrand.RandomState]] = None, n_jobs: int = - 1, silent: Union[bool, str] = 'warn', importance_type: str = 'split', **kwargs)[source]¶

PublicAPI (beta): This API is in beta and may change before becoming stable.

fit(X, y, sample_weight=None, init_score=None, eval_set=None, eval_names: Optional[List[str]] = None, eval_sample_weight=None, eval_init_score=None, eval_metric: Optional[Union[Callable, str, List[Union[Callable, str]]]] = None, ray_params: Union[None, lightgbm_ray.main.RayParams, Dict] = None, _remote: Optional[bool] = None, ray_dmatrix_params: Optional[Dict] = None, **kwargs: Any) → lightgbm_ray.sklearn.RayLGBMRegressor[source]¶

Build a gradient boosting model from the training set (X, y).

Parameters

X (array-like or sparse matrix of shape = [n_samples, n_features]) – Input feature matrix.
y (array-like of shape = [n_samples]) – The target values (class labels in classification, real numbers in regression).
sample_weight (array-like of shape = [n_samples] or None, optional (default=None)) – Weights of training data.
init_score (array-like of shape = [n_samples] or shape = [n_samples * n_classes] (for multi-class task) or shape = [n_samples, n_classes] (for multi-class task) or None, optional (default=None)) – Init score of training data.
eval_set (list or None, optional (default=None)) – A list of (X, y) tuple pairs to use as validation sets.
eval_names (list of str, or None, optional (default=None)) – Names of eval_set.
eval_sample_weight (list of array, or None, optional (default=None)) – Weights of eval data.
eval_init_score (list of array, or None, optional (default=None)) – Init score of eval data.
eval_metric (str, callable, list or None, optional (default=None)) – If str, it should be a built-in evaluation metric to use. If callable, it should be a custom evaluation metric, see note below for more details. If list, it can be a list of built-in metrics, a list of custom evaluation metrics, or a mix of both. In either case, the metric from the model parameters will be evaluated and used as well. Default: ‘l2’ for LGBMRegressor, ‘logloss’ for LGBMClassifier, ‘ndcg’ for LGBMRanker.
early_stopping_rounds (int or None, optional (default=None)) – Activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training. Requires at least one validation data and one metric. If there’s more than one, will check all of them. But the training data is ignored anyway. To check only the first metric, set the first_metric_only parameter to True in additional parameters **kwargs of the model constructor.
verbose (bool or int, optional (default=True)) –
Requires at least one evaluation data. If True, the eval metric on the eval set is printed at each boosting stage. If int, the eval metric on the eval set is printed at every verbose boosting stage. The last boosting stage or the boosting stage found by using early_stopping_rounds is also printed.

Example

With verbose = 4 and at least one item in eval_set, an evaluation metric is printed every 4 (instead of 1) boosting stages.
feature_name (list of str, or 'auto', optional (default='auto')) – Feature names. If ‘auto’ and data is pandas DataFrame, data columns names are used.
categorical_feature (list of str or int, or 'auto', optional (default='auto')) – Categorical features. If list of int, interpreted as indices. If list of str, interpreted as feature names (need to specify feature_name as well). If ‘auto’ and data is pandas DataFrame, pandas unordered categorical columns are used. All values in categorical features should be less than int32 max value (2147483647). Large values could be memory consuming. Consider using consecutive integers starting from zero. All negative values in categorical features will be treated as missing values. The output cannot be monotonically constrained with respect to a categorical feature.
callbacks (list of callable, or None, optional (default=None)) – List of callback functions that are applied at each iteration. See Callbacks in Python API for more information.
init_model (str, pathlib.Path, Booster, LGBMModel or None, optional (default=None)) – Filename of LightGBM model, Booster instance or LGBMModel instance used for continue training.
ray_params (lightgbm_ray.RayParams or dict, optional (default=None)) – Parameters to configure Ray-specific behavior. See lightgbm_ray.RayParams for a list of valid configuration parameters. Will override n_jobs attribute with own num_actors parameter.
_remote (bool, optional (default=False)) – Whether to run the driver process in a remote function. This is enabled by default in Ray client mode.
ray_dmatrix_params (dict, optional (default=None)) – Dict of parameters (such as sharding mode) passed to the internal RayDMatrix initialization.

Returns

self – Returns self.

Return type

object

Note

Custom eval function expects a callable with following signatures: func(y_true, y_pred), func(y_true, y_pred, weight) or func(y_true, y_pred, weight, group) and returns (eval_name, eval_result, is_higher_better) or list of (eval_name, eval_result, is_higher_better):

y_truearray-like of shape = [n_samples]
The target values.

y_predarray-like of shape = [n_samples] or shape = [n_samples * n_classes] (for multi-class task)
The predicted values. In case of custom objective, predicted values are returned before any transformation, e.g. they are raw margin instead of probability of positive class for binary task in this case.

weightarray-like of shape = [n_samples]
The weight of samples.

grouparray-like
Group/query data. Only used in the learning-to-rank task. sum(group) = n_samples. For example, if you have a 100-document dataset with group = [10, 20, 40, 10, 10, 10], that means that you have 6 groups, where the first 10 records are in the first group, records 11-30 are in the second group, records 31-70 are in the third group, etc.

eval_namestr
The name of evaluation function (without whitespace).

eval_resultfloat
The eval result.

is_higher_betterbool
Is eval result higher better, e.g. AUC is is_higher_better.

For multi-class task, the y_pred is group by class_id first, then group by row_id. If you want to get i-th row y_pred in j-th class, the access way is y_pred[j * num_data + i].

predict(X, *, ray_params: Union[None, lightgbm_ray.main.RayParams, Dict] = None, _remote: Optional[bool] = None, ray_dmatrix_params: Optional[Dict] = None, **kwargs)[source]¶

Return the predicted value for each sample.

Parameters

X (array-like or sparse matrix of shape = [n_samples, n_features]) – Input features matrix.
raw_score (bool, optional (default=False)) – Whether to predict raw scores.
start_iteration (int, optional (default=0)) – Start index of the iteration to predict. If <= 0, starts from the first iteration.
num_iteration (int or None, optional (default=None)) – Total number of iterations used in the prediction. If None, if the best iteration exists and start_iteration <= 0, the best iteration is used; otherwise, all iterations from start_iteration are used (no limits). If <= 0, all iterations from start_iteration are used (no limits).
pred_leaf (bool, optional (default=False)) – Whether to predict leaf index.
pred_contrib (bool, optional (default=False)) –
Whether to predict feature contributions.

Note

If you want to get more explanations for your model’s predictions using SHAP values, like SHAP interaction values, you can install the shap package (https://github.com/slundberg/shap). Note that unlike the shap package, with pred_contrib we return a matrix with an extra column, where the last column is the expected value.
ray_params (lightgbm_ray.RayParams or dict, optional (default=None)) – Parameters to configure Ray-specific behavior. See lightgbm_ray.RayParams for a list of valid configuration parameters. Will override n_jobs attribute with own num_actors parameter.
_remote (bool, optional (default=False)) – Whether to run the driver process in a remote function. This is enabled by default in Ray client mode.
ray_dmatrix_params (dict, optional (default=None)) – Dict of parameters (such as sharding mode) passed to the internal RayDMatrix initialization.
**kwargs – Other parameters for the prediction.

Returns

predicted_result (array-like of shape = [n_samples] or shape = [n_samples, n_classes]) – The predicted values.
X_leaves (array-like of shape = [n_samples, n_trees] or shape = [n_samples, n_trees * n_classes]) – If pred_leaf=True, the predicted leaf of every tree for each sample.
X_SHAP_values (array-like of shape = [n_samples, n_features + 1] or shape = [n_samples, (n_features + 1) * n_classes] or list with n_classes length of such objects) – If pred_contrib=True, the feature contributions for each sample.

to_local() → lightgbm.sklearn.LGBMRegressor[source]¶

Create regular version of lightgbm.LGBMRegressor from the distributed version.

Returns: model – Local underlying model.
Return type: lightgbm.LGBMRegressor

Ray 1.12.1

Distributed LightGBM on Ray¶

Installation¶

Usage¶

scikit-learn API¶

Data loading¶

Hyperparameter Tuning¶

Fault tolerance¶

Non-elastic training (warm restart)¶

Resources¶

Multi GPU training¶

How many remote actors should I use?¶

Distributed data loading¶

Data sources¶

Memory usage¶

Placement Strategies¶

More examples¶

API reference¶

scikit-learn API¶