grid_search.py

# -*- encoding: utf-8 -*-
from __future__ import division, print_function, absolute_import, unicode_literals

import itertools

import h2o
from h2o.job import H2OJob
from h2o.frame import H2OFrame
from h2o.estimators.estimator_base import H2OEstimator
from h2o.two_dim_table import H2OTwoDimTable
from h2o.display import H2ODisplay
from h2o.grid.metrics import *  # NOQA
from h2o.utils.backward_compatibility import backwards_compatible
from h2o.utils.shared_utils import deprecated, quoted
from h2o.utils.compatibility import *  # NOQA
from h2o.utils.typechecks import assert_is_type, is_type


class H2OGridSearch(backwards_compatible()):
    """
    Grid Search of a Hyper-Parameter Space for a Model

    :param model: The type of model to be explored initialized with optional parameters that will be
        unchanged across explored models.
    :param hyper_params: A dictionary of string parameters (keys) and a list of values to be explored by grid
        search (values).
    :param str grid_id: The unique id assigned to the resulting grid object. If none is given, an id will
        automatically be generated.
    :param search_criteria:  A dictionary of directives which control the search of the hyperparameter space.
        The default strategy "Cartesian" covers the entire space of hyperparameter combinations. Specify the
        "RandomDiscrete" strategy to get random search of all the combinations of your hyperparameters.
        RandomDiscrete should usually be combined with at least one early stopping criterion: max_models
        and/or max_runtime_secs, e.g::

            >>> criteria = {"strategy": "RandomDiscrete", "max_models": 42,
            ...             "max_runtime_secs": 28800, "seed": 1234}
            >>> criteria = {"strategy": "RandomDiscrete", "stopping_metric": "AUTO",
            ...             "stopping_tolerance": 0.001, "stopping_rounds": 10}
            >>> criteria = {"strategy": "RandomDiscrete", "stopping_rounds": 5,
            ...             "stopping_metric": "misclassification",
            ...             "stopping_tolerance": 0.00001}
    :returns: a new H2OGridSearch instance

    Examples
    --------
        >>> from h2o.grid.grid_search import H2OGridSearch
        >>> from h2o.estimators.glm import H2OGeneralizedLinearEstimator
        >>> hyper_parameters = {'alpha': [0.01,0.5], 'lambda': [1e-5,1e-6]}
        >>> gs = H2OGridSearch(H2OGeneralizedLinearEstimator(family='binomial'), hyper_parameters)
        >>> training_data = h2o.import_file("smalldata/logreg/benign.csv")
        >>> gs.train(x=range(3) + range(4,11),y=3, training_frame=training_data)
        >>> gs.show()
    """


    def __init__(self, model, hyper_params, grid_id=None, search_criteria=None):
        super(H2OGridSearch, self).__init__()
        assert_is_type(model, None, H2OEstimator, lambda mdl: issubclass(mdl, H2OEstimator))
        assert_is_type(hyper_params, dict)
        assert_is_type(grid_id, None, str)
        assert_is_type(search_criteria, None, dict)
        if not (model is None or is_type(model, H2OEstimator)): model = model()
        self._id = grid_id
        self.model = model
        self.hyper_params = dict(hyper_params)
        self.search_criteria = None if search_criteria is None else dict(search_criteria)
        self._grid_json = None
        self.models = None  # list of H2O Estimator instances
        self._parms = {}  # internal, for object recycle #
        self.parms = {}  # external#
        self._future = False  # used by __repr__/show to query job state#
        self._job = None  # used when _future is True#


    @property
    def grid_id(self):
        """A key that identifies this grid search object in H2O."""
        return self._id

    @grid_id.setter
    def grid_id(self, value):
        oldname = self.grid_id
        self._id = value
        h2o.rapids('(rename "{}" "{}")'.format(oldname, value))


    @property
    def model_ids(self):
        return [i['name'] for i in self._grid_json["model_ids"]]


    @property
    def hyper_names(self):
        return self._grid_json["hyper_names"]


    @property
    def failed_params(self):
        return self._grid_json.get("failed_params", None)


    @property
    def failure_details(self):
        return self._grid_json.get("failure_details", None)


    @property
    def failure_stack_traces(self):
        return self._grid_json.get("failure_stack_traces", None)


    @property
    def failed_raw_params(self):
        return self._grid_json.get("failed_raw_params", None)


    def start(self, x, y=None, training_frame=None, offset_column=None, fold_column=None, weights_column=None,
              validation_frame=None, **params):
        """
        Asynchronous model build by specifying the predictor columns, response column, and any
        additional frame-specific values.

        To block for results, call :meth:`join`.

        :param x: A list of column names or indices indicating the predictor columns.
        :param y: An index or a column name indicating the response column.
        :param training_frame: The H2OFrame having the columns indicated by x and y (as well as any
            additional columns specified by fold, offset, and weights).
        :param offset_column: The name or index of the column in training_frame that holds the offsets.
        :param fold_column: The name or index of the column in training_frame that holds the per-row fold
            assignments.
        :param weights_column: The name or index of the column in training_frame that holds the per-row weights.
        :param validation_frame: H2OFrame with validation data to be scored on while training.
        """
        self._future = True
        self.train(x=x,
                   y=y,
                   training_frame=training_frame,
                   offset_column=offset_column,
                   fold_column=fold_column,
                   weights_column=weights_column,
                   validation_frame=validation_frame,
                   **params)


    def join(self):
        """Wait until grid finishes computing."""
        self._future = False
        self._job.poll()
        self._job = None


    def train(self, x, y=None, training_frame=None, offset_column=None, fold_column=None, weights_column=None,
              validation_frame=None, **params):
        """
        Train the model synchronously (i.e. do not return until the model finishes training).

        To train asynchronously call :meth:`start`.

        :param x: A list of column names or indices indicating the predictor columns.
        :param y: An index or a column name indicating the response column.
        :param training_frame: The H2OFrame having the columns indicated by x and y (as well as any
            additional columns specified by fold, offset, and weights).
        :param offset_column: The name or index of the column in training_frame that holds the offsets.
        :param fold_column: The name or index of the column in training_frame that holds the per-row fold
            assignments.
        :param weights_column: The name or index of the column in training_frame that holds the per-row weights.
        :param validation_frame: H2OFrame with validation data to be scored on while training.
        """
        algo_params = locals()
        parms = self._parms.copy()
        parms.update({k: v for k, v in algo_params.items() if k not in ["self", "params", "algo_params", "parms"]})
        parms["search_criteria"] = self.search_criteria
        parms["hyper_parameters"] = self.hyper_params  # unique to grid search
        parms.update({k: v for k, v in list(self.model._parms.items()) if v is not None})  # unique to grid search
        parms.update(params)
        if '__class__' in parms:  # FIXME: hackt for PY3
            del parms['__class__']
        y = algo_params["y"]
        tframe = algo_params["training_frame"]
        if tframe is None: raise ValueError("Missing training_frame")
        if y is not None:
            if is_type(y, list, tuple):
                if len(y) == 1:
                    parms["y"] = y[0]
                else:
                    raise ValueError('y must be a single column reference')
            self._estimator_type = "classifier" if tframe[y].isfactor() else "regressor"
        self.build_model(parms)


    def build_model(self, algo_params):
        """(internal)"""
        if algo_params["training_frame"] is None: raise ValueError("Missing training_frame")
        x = algo_params.pop("x")
        y = algo_params.pop("y", None)
        training_frame = algo_params.pop("training_frame")
        validation_frame = algo_params.pop("validation_frame", None)
        is_auto_encoder = (algo_params is not None) and ("autoencoder" in algo_params and algo_params["autoencoder"])
        algo = self.model._compute_algo()  # unique to grid search
        is_unsupervised = is_auto_encoder or algo == "pca" or algo == "svd" or algo == "kmeans" or algo == "glrm"
        if is_auto_encoder and y is not None: raise ValueError("y should not be specified for autoencoder.")
        if not is_unsupervised and y is None: raise ValueError("Missing response")
        self._model_build(x, y, training_frame, validation_frame, algo_params)


    def _model_build(self, x, y, tframe, vframe, kwargs):
        kwargs['training_frame'] = tframe
        if vframe is not None: kwargs["validation_frame"] = vframe
        if is_type(y, int): y = tframe.names[y]
        if y is not None: kwargs['response_column'] = y
        if not is_type(x, list, tuple): x = [x]
        if is_type(x[0], int):
            x = [tframe.names[i] for i in x]
        offset = kwargs["offset_column"]
        folds = kwargs["fold_column"]
        weights = kwargs["weights_column"]
        ignored_columns = list(set(tframe.names) - set(x + [y, offset, folds, weights]))
        kwargs["ignored_columns"] = None if not ignored_columns else [quoted(col) for col in ignored_columns]
        kwargs = dict([(k, kwargs[k].frame_id if isinstance(kwargs[k], H2OFrame) else kwargs[k]) for k in kwargs if
                       kwargs[k] is not None])  # gruesome one-liner
        algo = self.model._compute_algo()  # unique to grid search
        if self.grid_id is not None: kwargs["grid_id"] = self.grid_id
        rest_ver = kwargs.pop("_rest_version") if "_rest_version" in kwargs else None

        grid = H2OJob(h2o.api("POST /99/Grid/%s" % algo, data=kwargs), job_type=(algo + " Grid Build"))

        if self._future:
            self._job = grid
            return

        grid.poll()
        if rest_ver is not None:
            grid_json = h2o.api("GET /99/Grids/%s" % (grid.dest_key))

            error_index = 0
            if len(grid_json["failure_details"]) > 0:
                print("Errors/Warnings building gridsearch model\n")

                for error_message in grid_json["failure_details"]:
                    if isinstance(grid_json["failed_params"][error_index], dict):
                        for h_name in grid_json['hyper_names']:
                            print("Hyper-parameter: {0}, {1}".format(h_name,
                                                                     grid_json['failed_params'][error_index][h_name]))

                    if len(grid_json["failure_stack_traces"]) > error_index:
                        print("failure_details: {0}\nfailure_stack_traces: "
                              "{1}\n".format(error_message, grid_json['failure_stack_traces'][error_index]))
                    error_index += 1
        else:
            grid_json = h2o.api("GET /99/Grids/%s" % grid.dest_key)

        self.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']]

        # get first model returned in list of models from grid search to get model class (binomial, multinomial, etc)
        # sometimes no model is returned due to bad parameter values provided by the user.
        if len(grid_json['model_ids']) > 0:
            first_model_json = h2o.api("GET /%d/Models/%s" %
                                       (rest_ver or 3, grid_json['model_ids'][0]['name']))['models'][0]
            self._resolve_grid(grid.dest_key, grid_json, first_model_json)
        else:
            raise ValueError("Gridsearch returns no model due to bad parameter values or other reasons....")


    def _resolve_grid(self, grid_id, grid_json, first_model_json):
        model_class = H2OGridSearch._metrics_class(first_model_json)
        m = model_class()
        m._id = grid_id
        m._grid_json = grid_json
        # m._metrics_class = metrics_class
        m._parms = self._parms
        H2OEstimator.mixin(self, model_class)
        self.__dict__.update(m.__dict__.copy())


    def __getitem__(self, item):
        return self.models[item]


    def __iter__(self):
        nmodels = len(self.models)
        return (self[i] for i in range(nmodels))


    def __len__(self):
        return len(self.models)


    def __repr__(self):
        self.show()
        return ""


    def predict(self, test_data):
        """
        Predict on a dataset.

        :param H2OFrame test_data: Data to be predicted on.
        :returns: H2OFrame filled with predictions.
        """
        return {model.model_id: model.predict(test_data) for model in self.models}


    def is_cross_validated(self):
        """Return True if the model was cross-validated."""
        return {model.model_id: model.is_cross_validated() for model in self.models}


    def xval_keys(self):
        """Model keys for the cross-validated model."""
        return {model.model_id: model.xval_keys() for model in self.models}


    def get_xval_models(self, key=None):
        """
        Return a Model object.

        :param str key: If None, return all cross-validated models; otherwise return the model
            specified by the key.
        :returns: A model or a list of models.
        """
        return {model.model_id: model.get_xval_models(key) for model in self.models}


    def xvals(self):
        """Return the list of cross-validated models."""
        return {model.model_id: model.xvals for model in self.models}


    def deepfeatures(self, test_data, layer):
        """
        Obtain a hidden layer's details on a dataset.

        :param test_data: Data to create a feature space on.
        :param int layer: Index of the hidden layer.
        :returns: A dictionary of hidden layer details for each model.
        """
        return {model.model_id: model.deepfeatures(test_data, layer) for model in self.models}


    def weights(self, matrix_id=0):
        """
        Return the frame for the respective weight matrix.

        :param: matrix_id: an integer, ranging from 0 to number of layers, that specifies the weight matrix to return.
        :returns: an H2OFrame which represents the weight matrix identified by matrix_id
        """
        return {model.model_id: model.weights(matrix_id) for model in self.models}


    def biases(self, vector_id=0):
        """
        Return the frame for the respective bias vector.

        :param: vector_id: an integer, ranging from 0 to number of layers, that specifies the bias vector to return.
        :returns: an H2OFrame which represents the bias vector identified by vector_id
        """
        return {model.model_id: model.biases(vector_id) for model in self.models}


    def normmul(self):
        """Normalization/Standardization multipliers for numeric predictors."""
        return {model.model_id: model.normmul() for model in self.models}


    def normsub(self):
        """Normalization/Standardization offsets for numeric predictors."""
        return {model.model_id: model.normsub() for model in self.models}


    def respmul(self):
        """Normalization/Standardization multipliers for numeric response."""
        return {model.model_id: model.respmul() for model in self.models}


    def respsub(self):
        """Normalization/Standardization offsets for numeric response."""
        return {model.model_id: model.respsub() for model in self.models}


    def catoffsets(self):
        """
        Categorical offsets for one-hot encoding
        """
        return {model.model_id: model.catoffsets() for model in self.models}


    def model_performance(self, test_data=None, train=False, valid=False, xval=False):
        """
        Generate model metrics for this model on test_data.

        :param test_data: Data set for which model metrics shall be computed against. All three of train, valid
            and xval arguments are ignored if test_data is not None.
        :param train: Report the training metrics for the model.
        :param valid: Report the validation metrics for the model.
        :param xval: Report the validation metrics for the model.
        :return: An object of class H2OModelMetrics.
        """
        return {model.model_id: model.model_performance(test_data, train, valid, xval) for model in self.models}


    def scoring_history(self):
        """
        Retrieve model scoring history.

        :returns: Score history (H2OTwoDimTable)
        """
        return {model.model_id: model.scoring_history() for model in self.models}


    def summary(self, header=True):
        """Print a detailed summary of the explored models."""
        table = []
        for model in self.models:
            model_summary = model._model_json["output"]["model_summary"]
            r_values = list(model_summary.cell_values[0])
            r_values[0] = model.model_id
            table.append(r_values)

        # if h2o.can_use_pandas():
        #  import pandas
        #  pandas.options.display.max_rows = 20
        #  print pandas.DataFrame(table,columns=self.col_header)
        #  return
        print()
        if header:
            print('Grid Summary:')
        print()
        H2ODisplay(table, ['Model Id'] + model_summary.col_header[1:], numalign="left", stralign="left")


    def show(self):
        """Print models sorted by metric."""
        hyper_combos = itertools.product(*list(self.hyper_params.values()))
        if not self.models:
            c_values = [[idx + 1, list(val)] for idx, val in enumerate(hyper_combos)]
            print(H2OTwoDimTable(
                col_header=['Model', 'Hyperparameters: [' + ', '.join(list(self.hyper_params.keys())) + ']'],
                table_header='Grid Search of Model ' + self.model.__class__.__name__, cell_values=c_values))
        else:
            print(self.sorted_metric_table())


    def varimp(self, use_pandas=False):
        """
        Pretty print the variable importances, or return them in a list/pandas DataFrame.

        :param bool use_pandas: If True, then the variable importances will be returned as a pandas data frame.

        :returns: A dictionary of lists or Pandas DataFrame instances.
        """
        return {model.model_id: model.varimp(use_pandas) for model in self.models}


    def residual_deviance(self, train=False, valid=False, xval=False):
        """
        Retreive the residual deviance if this model has the attribute, or None otherwise.

        :param bool train: Get the residual deviance for the training set. If both train and valid are False,
            then train is selected by default.
        :param bool valid: Get the residual deviance for the validation set. If both train and valid are True,
            then train is selected by default.
        :param bool xval: Get the residual deviance for the cross-validated models.

        :returns: the residual deviance, or None if it is not present.
        """
        return {model.model_id: model.residual_deviance(train, valid, xval) for model in self.models}


    def residual_degrees_of_freedom(self, train=False, valid=False, xval=False):
        """
        Retreive the residual degress of freedom if this model has the attribute, or None otherwise.

        :param bool train: Get the residual dof for the training set. If both train and valid are False, then
            train is selected by default.
        :param bool valid: Get the residual dof for the validation set. If both train and valid are True, then
            train is selected by default.
        :param bool xval: Get the residual dof for the cross-validated models.

        :returns: the residual degrees of freedom, or None if they are not present.
        """
        return {model.model_id: model.residual_degrees_of_freedom(train, valid, xval) for model in self.models}


    def null_deviance(self, train=False, valid=False, xval=False):
        """
        Retreive the null deviance if this model has the attribute, or None otherwise.

        :param bool train: Get the null deviance for the training set. If both train and valid are False, then
            train is selected by default.
        :param bool valid: Get the null deviance for the validation set. If both train and valid are True, then
            train is selected by default.
        :param bool xval: Get the null deviance for the cross-validated models.

        :returns: the null deviance, or None if it is not present.
        """
        return {model.model_id: model.null_deviance(train, valid, xval) for model in self.models}


    def null_degrees_of_freedom(self, train=False, valid=False, xval=False):
        """
        Retreive the null degress of freedom if this model has the attribute, or None otherwise.

        :param bool train: Get the null dof for the training set. If both train and valid are False, then train is
            selected by default.
        :param bool valid: Get the null dof for the validation set. If both train and valid are True, then train is
            selected by default.
        :param bool xval: Get the null dof for the cross-validated models.

        :returns: the null dof, or None if it is not present.
        """
        return {model.model_id: model.null_degrees_of_freedom(train, valid, xval) for model in self.models}


    def pprint_coef(self):
        """Pretty print the coefficents table (includes normalized coefficients)."""
        for i, model in enumerate(self.models):
            print('Model', i)
            model.pprint_coef()
            print()


    def coef(self):
        """Return the coefficients for this model."""
        return {model.model_id: model.coef() for model in self.models}


    def coef_norm(self):
        """Return the normalized coefficients."""
        return {model.model_id: model.coef_norm() for model in self.models}


    def r2(self, train=False, valid=False, xval=False):
        """
        Return the R^2 for this regression model.

        The R^2 value is defined to be ``1 - MSE/var``, where ``var`` is computed as ``sigma^2``.

        If all are False (default), then return the training metric value.
        If more than one options is set to True, then return a dictionary of metrics where the keys are "train",
        "valid", and "xval".

        :param bool train: If train is True, then return the R^2 value for the training data.
        :param bool valid: If valid is True, then return the R^2 value for the validation data.
        :param bool xval:  If xval is True, then return the R^2 value for the cross validation data.

        :returns: The R^2 for this regression model.
        """
        return {model.model_id: model.r2(train, valid, xval) for model in self.models}


    def mse(self, train=False, valid=False, xval=False):
        """
        Get the MSE(s).

        If all are False (default), then return the training metric value.
        If more than one options is set to True, then return a dictionary of metrics where the keys are "train",
        "valid", and "xval".

        :param bool train: If train is True, then return the MSE value for the training data.
        :param bool valid: If valid is True, then return the MSE value for the validation data.
        :param bool xval:  If xval is True, then return the MSE value for the cross validation data.
        :returns: The MSE for this regression model.
        """
        return {model.model_id: model.mse(train, valid, xval) for model in self.models}


    def logloss(self, train=False, valid=False, xval=False):
        """
        Get the Log Loss(s).

        If all are False (default), then return the training metric value.
        If more than one options is set to True, then return a dictionary of metrics where the keys are "train",
        "valid", and "xval".

        :param bool train: If train is True, then return the Log Loss value for the training data.
        :param bool valid: If valid is True, then return the Log Loss value for the validation data.
        :param bool xval:  If xval is True, then return the Log Loss value for the cross validation data.

        :returns: The Log Loss for this binomial model.
        """
        return {model.model_id: model.logloss(train, valid, xval) for model in self.models}


    def mean_residual_deviance(self, train=False, valid=False, xval=False):
        """
        Get the Mean Residual Deviances(s).

        If all are False (default), then return the training metric value.
        If more than one options is set to True, then return a dictionary of metrics where the keys are "train",
        "valid", and "xval".

        :param bool train: If train is True, then return the Mean Residual Deviance value for the training data.
        :param bool valid: If valid is True, then return the Mean Residual Deviance value for the validation data.
        :param bool xval:  If xval is True, then return the Mean Residual Deviance value for the cross validation data.
        :returns: The Mean Residual Deviance for this regression model.
        """
        return {model.model_id: model.mean_residual_deviance(train, valid, xval) for model in self.models}


    def auc(self, train=False, valid=False, xval=False):
        """
        Get the AUC(s).

        If all are False (default), then return the training metric value.
        If more than one options is set to True, then return a dictionary of metrics where the keys are "train",
        "valid", and "xval".

        :param bool train: If train is True, then return the AUC value for the training data.
        :param bool valid: If valid is True, then return the AUC value for the validation data.
        :param bool xval:  If xval is True, then return the AUC value for the validation data.

        :returns: The AUC.
        """
        return {model.model_id: model.auc(train, valid, xval) for model in self.models}


    def aic(self, train=False, valid=False, xval=False):
        """
        Get the AIC(s).

        If all are False (default), then return the training metric value.
        If more than one options is set to True, then return a dictionary of metrics where the keys are "train",
        "valid", and "xval".

        :param bool train: If train is True, then return the AIC value for the training data.
        :param bool valid: If valid is True, then return the AIC value for the validation data.
        :param bool xval:  If xval is True, then return the AIC value for the validation data.

        :returns: The AIC.
        """
        return {model.model_id: model.aic(train, valid, xval) for model in self.models}


    def gini(self, train=False, valid=False, xval=False):
        """
        Get the Gini Coefficient(s).

        If all are False (default), then return the training metric value.
        If more than one options is set to True, then return a dictionary of metrics where the keys are "train",
        "valid", and "xval".

        :param bool train: If train is True, then return the Gini Coefficient value for the training data.
        :param bool valid: If valid is True, then return the Gini Coefficient value for the validation data.
        :param bool xval:  If xval is True, then return the Gini Coefficient value for the cross validation data.

        :returns: The Gini Coefficient for this binomial model.
        """
        return {model.model_id: model.gini(train, valid, xval) for model in self.models}


    def get_hyperparams(self, id, display=True):
        """
        Get the hyperparameters of a model explored by grid search.

        :param str id: The model id of the model with hyperparameters of interest.
        :param bool display: Flag to indicate whether to display the hyperparameter names.

        :returns: A list of the hyperparameters for the specified model.
        """
        idx = id if is_type(id, int) else self.model_ids.index(id)
        model = self[idx]

        # if cross-validation is turned on, parameters in one of the fold model actuall contains the max_runtime_secs
        # parameter and not the main model that is returned.
        if model._is_xvalidated:
            model = h2o.get_model(model._xval_keys[0])

        res = [model.params[h]['actual'][0] if isinstance(model.params[h]['actual'], list)
               else model.params[h]['actual']
               for h in self.hyper_params]
        if display: print('Hyperparameters: [' + ', '.join(list(self.hyper_params.keys())) + ']')
        return res


    def get_hyperparams_dict(self, id, display=True):
        """
        Derived and returned the model parameters used to train the particular grid search model.

        :param str id: The model id of the model with hyperparameters of interest.
        :param bool display: Flag to indicate whether to display the hyperparameter names.

        :returns: A dict of model pararmeters derived from the hyper-parameters used to train this particular model.
        """
        idx = id if is_type(id, int) else self.model_ids.index(id)
        model = self[idx]

        model_params = dict()

        # if cross-validation is turned on, parameters in one of the fold model actual contains the max_runtime_secs
        # parameter and not the main model that is returned.
        if model._is_xvalidated:
            model = h2o.get_model(model._xval_keys[0])

        for param_name in self.hyper_names:
            model_params[param_name] = model.params[param_name]['actual'][0] if \
                isinstance(model.params[param_name]['actual'], list) else model.params[param_name]['actual']

        if display: print('Hyperparameters: [' + ', '.join(list(self.hyper_params.keys())) + ']')
        return model_params


    def sorted_metric_table(self):
        """
        Retrieve summary table of an H2O Grid Search.

        :returns: The summary table as an H2OTwoDimTable or a Pandas DataFrame.
        """
        summary = self._grid_json["summary_table"]
        if summary is not None: return summary.as_data_frame()
        print("No sorted metric table for this grid search")


    @staticmethod
    def _metrics_class(model_json):
        model_type = model_json["output"]["model_category"]
        if model_type == "Binomial":
            model_class = H2OBinomialGridSearch
        elif model_type == "Clustering":
            model_class = H2OClusteringGridSearch
        elif model_type == "Regression":
            model_class = H2ORegressionGridSearch
        elif model_type == "Multinomial":
            model_class = H2OMultinomialGridSearch
        elif model_type == "AutoEncoder":
            model_class = H2OAutoEncoderGridSearch
        elif model_type == "DimReduction":
            model_class = H2ODimReductionGridSearch
        else:
            raise NotImplementedError(model_type)
        return model_class


    def get_grid(self, sort_by=None, decreasing=None):
        """
        Retrieve an H2OGridSearch instance.

        Optionally specify a metric by which to sort models and a sort order.
        Note that if neither cross-validation nor a validation frame is used in the grid search, then the
        training metrics will display in the "get grid" output. If a validation frame is passed to the grid, and
        ``nfolds = 0``, then the validation metrics will display. However, if ``nfolds`` > 1, then cross-validation
        metrics will display even if a validation frame is provided.

        :param str sort_by: A metric by which to sort the models in the grid space. Choices are: ``"logloss"``,
            ``"residual_deviance"``, ``"mse"``, ``"auc"``, ``"r2"``, ``"accuracy"``, ``"precision"``, ``"recall"``,
            ``"f1"``, etc.
        :param bool decreasing: Sort the models in decreasing order of metric if true, otherwise sort in increasing
            order (default).

        :returns: A new H2OGridSearch instance optionally sorted on the specified metric.
        """
        if sort_by is None and decreasing is None: return self

        grid_json = h2o.api("GET /99/Grids/%s" % self._id, data={"sort_by": sort_by, "decreasing": decreasing})
        grid = H2OGridSearch(self.model, self.hyper_params, self._id)
        grid.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']]  # reordered
        first_model_json = h2o.api("GET /99/Models/%s" % grid_json['model_ids'][0]['name'])['models'][0]
        model_class = H2OGridSearch._metrics_class(first_model_json)
        m = model_class()
        m._id = self._id
        m._grid_json = grid_json
        # m._metrics_class = metrics_class
        m._parms = grid._parms
        H2OEstimator.mixin(grid, model_class)
        grid.__dict__.update(m.__dict__.copy())
        return grid


    # Deprecated functions; left here for backward compatibility
    _bcim = {
        "giniCoef": lambda self, *args, **kwargs: self.gini(*args, **kwargs)
    }


    @deprecated("grid.sort_by() is deprecated; use grid.get_grid() instead")
    def sort_by(self, metric, increasing=True):
        """Deprecated since 2016-12-12, use grid.get_grid() instead."""

        if metric[-1] != ')': metric += '()'
        c_values = [list(x) for x in zip(*sorted(eval('self.' + metric + '.items()'), key=lambda k_v: k_v[1]))]
        c_values.insert(1, [self.get_hyperparams(model_id, display=False) for model_id in c_values[0]])
        if not increasing:
            for col in c_values: col.reverse()
        if metric[-2] == '(': metric = metric[:-2]
        return H2OTwoDimTable(
            col_header=['Model Id', 'Hyperparameters: [' + ', '.join(list(self.hyper_params.keys())) + ']', metric],
            table_header='Grid Search Results for ' + self.model.__class__.__name__,
            cell_values=[list(x) for x in zip(*c_values)])