engine.py

"""Training Library containing training routines of LightGBM."""
from __future__ import absolute_import

import collections
import numpy as np
from .basic import LightGBMError, Predictor, Dataset, Booster, is_str
from . import callback

def _construct_dataset(data, reference=None,
    params=None, other_fields=None, predictor=None):
    if 'max_bin' in params:
        max_bin = int(params['max_bin'])
    else:
        max_bin = 255
    weight = None
    group = None
    init_score = None
    if other_fields is not None:
        if not is isinstance(other_fields, dict):
            raise TypeError("other filed data should be dict type")
        weight = None if 'weight' not in other_fields else other_fields['weight']
        group = None if 'group' not in other_fields else other_fields['group']
        init_score = None if 'init_score' not in other_fields else other_fields['init_score']
    if reference is None:
        if is_str(data):
            ret = Dataset(data, label=None, max_bin=max_bin, 
            weight=weight, group=group, predictor=predictor, params=params)
        else:
            ret = Dataset(data[0], data[1], max_bin=max_bin, 
                weight=weight, group=group, predictor=predictor, params=params)
    else:
        if is_str(data):
            ret = reference.create_valid(data, label=None, weight=weight, group=group, params=params)
        else:
            ret = reference.create_valid(data[0], data[1], weight, group, params=params)
    if init_score is not None:
        ret.set_init_score(init_score)
    return ret

def train(params, train_data, num_boost_round=100, 
        valid_datas=None, valid_names=None,
        fobj=None, feval=None, init_model=None, 
        train_fields=None, valid_fields=None, 
        early_stopping_rounds=None, out_eval_result=None,
        verbose_eval=True, learning_rates=None, callbacks=None):
    """Train with given parameters.

    Parameters
    ----------
    params : dict
         params.
    train_data : pair, (X, y) or filename of data
        Data to be trained.
    num_boost_round: int
        Number of boosting iterations.
    valid_datas: list of pairs (valid_X, valid_y) or filename of data
        List of data to be evaluated during training
    valid_names: list of string
        names of valid_datas
    fobj : function
        Customized objective function.
    feval : function
        Customized evaluation function.
        Note: should return (eval_name, eval_result, is_higher_better) of list of this
    init_model : file name of lightgbm model or 'Booster' instance
        model used for continued train
    train_fields : dict
        other data file in training data. e.g. train_fields['weight'] is weight data
        support fields: weight, group, init_score
    valid_fields : dict
        other data file in training data. e.g. valid_fields[0]['weight'] is weight data for first valid data
        support fields: weight, group, init_score
    early_stopping_rounds: int
        Activates early stopping. 
        Requires at least one validation data and one metric
        If there's more than one, will check all of them
        Returns the model with (best_iter + early_stopping_rounds)
        If early stopping occurs, the model will add 'best_iteration' field
    out_eval_result: dict or None
        This dictionary used to store all evaluation results of all the items in valid_datas.
        Example: with a valid_datas containing [valid_set, train_set] and valid_names containing ['eval', 'train'] and
        a paramater containing ('metric':'logloss')
        Returns: {'train': {'logloss': ['0.48253', '0.35953', ...]},
                  'eval': {'logloss': ['0.480385', '0.357756', ...]}}
        passed with None means no using this function
    verbose_eval : bool or int
        Requires at least one item in evals.
        If `verbose_eval` is True then the evaluation metric on the validation set is
        printed at each boosting stage.
        If `verbose_eval` is an integer then the evaluation metric on the validation set
        is printed at every given `verbose_eval` boosting stage. The last boosting stage
        / the boosting stage found by using `early_stopping_rounds` is also printed.
        Example: with verbose_eval=4 and at least one item in evals, an evaluation metric
        is printed every 4 boosting stages, instead of every boosting stage.
    learning_rates: list or function
        List of learning rate for each boosting round
        or a customized function that calculates learning_rate in terms of
        current number of round and the total number of boosting round (e.g. yields
        learning rate decay)
        - list l: learning_rate = l[current_round]
        - function f: learning_rate = f(current_round, total_boost_round)
    callbacks : list of callback functions
        List of callback functions that are applied at end of each iteration.

    Returns
    -------
    booster : a trained booster model
    """
    """create predictor first"""
    if is_str(init_model):
        predictor = Predictor(model_file=init_model)
    elif isinstance(init_model, Booster):
        predictor = Booster.to_predictor()
    elif isinstance(init_model, Predictor):
        predictor = init_model
    else:
        predictor = None
    """create dataset"""
    train_set = _construct_dataset(train_data, None, params, train_fields, predictor)
    is_valid_contain_train = False
    train_data_name = "training"
    valid_sets = []
    name_valid_sets = []
    if valid_datas is not None:
        for i in range(len(valid_datas)):
            other_fields = None if valid_fields is None else valid_fields[i]
            """reduce cost for prediction training data"""
            if valid_datas[i] is train_data:
                is_valid_contain_train = True
                train_data_name = valid_names[i]
                continue
            valid_set = _construct_dataset(
                valid_datas[i],
                train_set, 
                params, 
                other_fields, 
                predictor)
            valid_sets.append(valid_set)
            name_valid_sets.append(valid_names[i])
    """process callbacks"""
    callbacks = [] if callbacks is None else callbacks

    # Most of legacy advanced options becomes callbacks
    if isinstance(verbose_eval, bool) and verbose_eval:
        callbacks.append(callback.print_evaluation())
    else:
        if isinstance(verbose_eval, int):
            callbacks.append(callback.print_evaluation(verbose_eval))

    if early_stopping_rounds is not None:
        callbacks.append(callback.early_stop(early_stopping_rounds,
                                             verbose=bool(verbose_eval)))
    if learning_rates is not None:
        callbacks.append(callback.reset_learning_rate(learning_rates))

    if evals_result is not None:
        callbacks.append(callback.record_evaluation(evals_result))

    callbacks_before_iter = [
        cb for cb in callbacks if cb.__dict__.get('before_iteration', False)]
    callbacks_after_iter = [
        cb for cb in callbacks if not cb.__dict__.get('before_iteration', False)]
    """construct booster"""
    booster = Booster(params=params, train_set=train_set)
    if is_valid_contain_train:
        booster.set_train_data_name(train_data_name)
    for i in range(len(valid_sets)):
        booster.add_valid(valid_sets[i], name_valid_sets[i])
    """start training"""
    for i in range(num_boost_round):
        for cb in callbacks_before_iter:
            cb(callback.CallbackEnv(model=booster,
                           cvfolds=None,
                           iteration=i,
                           begin_iteration=0,
                           end_iteration=num_boost_round,
                           evaluation_result_list=None))

        booster.update(fobj=fobj)

        evaluation_result_list = []
        # check evaluation result.
        if len(valid_sets) != 0:
            if is_valid_contain_train:
                evaluation_result_list.extend(booster.eval_train(feval))
            evaluation_result_list.extend(booster.eval_valid(feval))
        try:
            for cb in callbacks_after_iter:
                cb(callback.CallbackEnv(model=booster,
                               cvfolds=None,
                               iteration=i,
                               begin_iteration=0,
                               end_iteration=num_boost_round,
                               evaluation_result_list=evaluation_result_list))
        except callback.EarlyStopException:
            break
    if booster.attr('best_iteration') is not None:
        booster.best_iteration = int(booster.attr('best_iteration'))
    else:
        booster.best_iteration = num_boost_round - 1
    return num_boost_round


class CVBooster(object):
    """"Auxiliary datastruct to hold one fold of CV."""
    def __init__(self, train_set, valid_test, param):
        """"Initialize the CVBooster"""
        self.train_set = train_set
        self.valid_test = valid_test
        self.booster = Booster(params=params, train_set=train_set)
        self.booster.add_valid(valid_test, 'valid')

    def update(self, fobj):
        """"Update the boosters for one iteration"""
        self.booster.update(fobj=fobj)

    def eval(self, feval):
        """"Evaluate the CVBooster for one iteration."""
        return self.booster.eval_valid(feval)

try:
    try:
        from sklearn.model_selection import KFold, StratifiedKFold
    except ImportError:
        from sklearn.cross_validation import KFold, StratifiedKFold
    SKLEARN_StratifiedKFold = True
except ImportError:
    SKLEARN_StratifiedKFold = False

def _make_n_folds(full_data, nfold, param, seed, fpreproc=None, stratified=False):
    """
    Make an n-fold list of CVBooster from random indices.
    """
    np.random.seed(seed)
    if stratified:
        if SKLEARN_StratifiedKFold:
            sfk = StratifiedKFold(n_splits=nfold, shuffle=True, random_state=seed)
            idset = [x[1] for x in sfk.split(X=full_data.get_label(), y=full_data.get_label())]
        else:
            raise LightGBMError('sklearn needs to be installed in order to use stratified cv')
    else:
        randidx = np.random.permutation(full_data.num_data())
        kstep = int(len(randidx) / nfold)
        idset = [randidx[(i * kstep): min(len(randidx), (i + 1) * kstep)] for i in range(nfold)]

    ret = []
    for k in range(nfold):
        train_set = full_data.subset(np.concatenate([idset[i] for i in range(nfold) if k != i]))
        valid_set = full_data.subset(idset[k])
        # run preprocessing on the data set if needed
        if fpreproc is not None:
            train_set, valid_set, tparam = fpreproc(train_set, valid_set, param.copy())
        else:
            tparam = param
        ret.append(CVBooster(train_set, valid_set, tparam))
    return ret

def _agg_cv_result(raw_results):
    # pylint: disable=invalid-name
    """
    Aggregate cross-validation results.
    """
    cvmap = {}
    metric_type = {}
    for one_result in raw_results:
        for one_line in one_result:
            key = one_line[1]
            metric_type[key] = one_line[3]
            if key not in cvmap:
                cvmap[key] = []
            cvmap[key].append(one_result[2])
    results = []
    for k, v in cvmap.items():
        v = np.array(v)
        mean, std = np.mean(v), np.std(v)
        results.extend(['cv_agg', k, mean, metric_type[k], std])
    return results

def cv(params, train_data, num_boost_round=10, nfold=5, stratified=False,
       metrics=(), fobj=None, feval=None, train_fields=None, early_stopping_rounds=None,
       fpreproc=None, verbose_eval=None, show_stdv=True, seed=0,
       callbacks=None):
    # pylint: disable = invalid-name
    """Cross-validation with given paramaters.

    Parameters
    ----------
    params : dict
        Booster params.
    train_data : pair, (X, y) or filename of data
        Data to be trained.
    num_boost_round : int
        Number of boosting iterations.
    nfold : int
        Number of folds in CV.
    stratified : bool
        Perform stratified sampling.
    folds : a KFold or StratifiedKFold instance
        Sklearn KFolds or StratifiedKFolds.
    metrics : string or list of strings
        Evaluation metrics to be watched in CV.
    fobj : function
        Custom objective function.
    feval : function
        Custom evaluation function.
    train_fields : dict
        other data file in training data. e.g. train_fields['weight'] is weight data
        support fields: weight, group, init_score
    early_stopping_rounds: int
        Activates early stopping. CV error needs to decrease at least
        every <early_stopping_rounds> round(s) to continue.
        Last entry in evaluation history is the one from best iteration.
    fpreproc : function
        Preprocessing function that takes (dtrain, dtest, param) and returns
        transformed versions of those.
    verbose_eval : bool, int, or None, default None
        Whether to display the progress. If None, progress will be displayed
        when np.ndarray is returned. If True, progress will be displayed at
        boosting stage. If an integer is given, progress will be displayed
        at every given `verbose_eval` boosting stage.
    show_stdv : bool, default True
        Whether to display the standard deviation in progress.
        Results are not affected, and always contains std.
    seed : int
        Seed used to generate the folds (passed to numpy.random.seed).
    callbacks : list of callback functions
        List of callback functions that are applied at end of each iteration.

    Returns
    -------
    evaluation history : list(string)
    """

    if isinstance(metrics, str):
        metrics = [metrics]

    if isinstance(params, list):
        params = dict(params)

    if not 'metric' in params:
        params['metric'] = []

    if len(metric) > 0:
        params['metric'].extend(metric)

    train_set = _construct_dataset(train_data, None, params, train_fields)

    results = {}
    cvfolds = _make_n_folds(train_set, nfold, params, seed, fpreproc, stratified)

    # setup callbacks
    callbacks = [] if callbacks is None else callbacks
    if early_stopping_rounds is not None:
        callbacks.append(callback.early_stop(early_stopping_rounds,
                                             verbose=False))
    if isinstance(verbose_eval, bool) and verbose_eval:
        callbacks.append(callback.print_evaluation(show_stdv=show_stdv))
    else:
        if isinstance(verbose_eval, int):
            callbacks.append(callback.print_evaluation(verbose_eval, show_stdv=show_stdv))

    callbacks_before_iter = [
        cb for cb in callbacks if cb.__dict__.get('before_iteration', False)]
    callbacks_after_iter = [
        cb for cb in callbacks if not cb.__dict__.get('before_iteration', False)]

    for i in range(num_boost_round):
        for cb in callbacks_before_iter:
            cb(callback.CallbackEnv(model=None,
                           cvfolds=cvfolds,
                           iteration=i,
                           begin_iteration=0,
                           end_iteration=num_boost_round,
                           evaluation_result_list=None))
        for fold in cvfolds:
            fold.update(fobj)
        res = aggcv([f.eval(feval) for f in cvfolds])

        for _, key, mean, _, std in res:
            if key + '-mean' not in results:
                results[key + '-mean'] = []
            if key + '-std' not in results:
                results[key + '-std'] = []
            results[key + '-mean'].append(mean)
            results[key + '-std'].append(std)
        try:
            for cb in callbacks_after_iter:
                cb(callback.CallbackEnv(model=None,
                               cvfolds=cvfolds,
                               iteration=i,
                               begin_iteration=0,
                               end_iteration=num_boost_round,
                               evaluation_result_list=res))
        except callback.EarlyStopException as e:
            for k in results.keys():
                results[k] = results[k][:(e.final_best_iter + 1)]
            break
    return results