In [3]:
# Import potrebných balíčkov
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, OrdinalEncoder
from sklearn.impute import SimpleImputer
from sklearn.compose import make_column_transformer
from sklearn.pipeline import make_pipeline
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score, precision_score, recall_score

from hyperopt import hp, tpe
from hyperopt.pyll.base import scope
from hyperopt.fmin import fmin
from hyperopt import space_eval
from sklearn.model_selection import cross_validate
In [4]:
# Uistíme sa, že máme všetky potrebné dáta
!mkdir -p data/titanic
!wget -nc -O data/titanic.zip https://www.dropbox.com/s/u8u7vcwy3sosbar/titanic.zip?dl=1
!unzip -oq -d data/titanic data/titanic.zip
--2019-08-26 23:13:34--  https://www.dropbox.com/s/u8u7vcwy3sosbar/titanic.zip?dl=1
Resolving www.dropbox.com (www.dropbox.com)... 162.125.66.1, 2620:100:6027:1::a27d:4801
Connecting to www.dropbox.com (www.dropbox.com)|162.125.66.1|:443... connected.
HTTP request sent, awaiting response... 301 Moved Permanently
Location: /s/dl/u8u7vcwy3sosbar/titanic.zip [following]
--2019-08-26 23:13:35--  https://www.dropbox.com/s/dl/u8u7vcwy3sosbar/titanic.zip
Reusing existing connection to www.dropbox.com:443.
HTTP request sent, awaiting response... 302 Found
Location: https://uc7781c3d98736df4aed81c0d69f.dl.dropboxusercontent.com/cd/0/get/AnbwilOVeiaWGYj9dz1Fb5wuafNtJFy2QMpPr-z-bBT2sFaah6vhH5d5SfiZv87UHRov0yRgAeWwb5bkzntt5lOruzQ9ToPaFFmdOa_ap7UHaNokaL7Y_QP-m2crRH4fAVA/file?dl=1# [following]
--2019-08-26 23:13:35--  https://uc7781c3d98736df4aed81c0d69f.dl.dropboxusercontent.com/cd/0/get/AnbwilOVeiaWGYj9dz1Fb5wuafNtJFy2QMpPr-z-bBT2sFaah6vhH5d5SfiZv87UHRov0yRgAeWwb5bkzntt5lOruzQ9ToPaFFmdOa_ap7UHaNokaL7Y_QP-m2crRH4fAVA/file?dl=1
Resolving uc7781c3d98736df4aed81c0d69f.dl.dropboxusercontent.com (uc7781c3d98736df4aed81c0d69f.dl.dropboxusercontent.com)... 162.125.66.6, 2620:100:6022:6::a27d:4206
Connecting to uc7781c3d98736df4aed81c0d69f.dl.dropboxusercontent.com (uc7781c3d98736df4aed81c0d69f.dl.dropboxusercontent.com)|162.125.66.6|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 34303 (33K) [application/binary]
Saving to: ‘data/titanic.zip’

data/titanic.zip    100%[===================>]  33,50K  --.-KB/s    in 0,05s   

2019-08-26 23:13:36 (635 KB/s) - ‘data/titanic.zip’ saved [34303/34303]

Bayesovská optimalizácia hyperparametrov

Keď už vieme, ako sa dá pomocou balíčka scikit-learn a pipelines natrénovať jednoduchý model, môžeme sa venovať téme optimalizácie hyperparametrov. Ako vieme, v podstate každá metóda strojového učenia má určité nastaviteľné konštanty, ktoré čiastočne determinujú jej správanie a ktoré nazývame hyperparametre.

Nastavenie hyperparametrov môže mať kľúčový dopad na to, ako dobre bude metóda fungovať. Navyše – pre každú úlohu môžu byť potrebné trochu iné hyperparametre. Preto je dobré poznať účinné metódy, ako hyperparametre naladiť: buď ručne alebo ešte lepšie automaticky.

Jednej takej metóde — bayesovskej optimalizácii – sa venuje tento notebook.

Načítanie dát a predspracovanie

Ako obvykle, začneme načítaním a predspracovaním dát. Budeme znovu pracovať s dátovou množinou Titanic, ktorú sme používali v predchádzajúcom príklade. Načítanie a predspracovanie dát na tejto dátovej množine sme si už precvičili, preto to nebudeme robiť znovu v tomto notebook-u a použijeme predpripravený kód:

In [5]:
df = pd.read_csv("data/titanic/train.csv")
df_train, df_test = train_test_split(df, test_size=0.25,
                     stratify=df["Survived"], random_state=4)

categorical_inputs = ["Pclass", "Sex", "Embarked"]
numeric_inputs = ["Age", "SibSp", 'Parch', 'Fare']
output = ["Survived"]

X_train = df_train[categorical_inputs + numeric_inputs]
Y_train = df_train[output]

X_test = df_test[categorical_inputs + numeric_inputs]
Y_test = df_test[output]


input_preproc = make_column_transformer(
    (make_pipeline(SimpleImputer(strategy='constant', fill_value='MISSING'),
                   OrdinalEncoder()), categorical_inputs),
    (make_pipeline(SimpleImputer(), StandardScaler()), numeric_inputs)
)

Vstupné tréningové dáta si tento raz hneď aj predspracujeme. Budeme na nich trénovať veľa rôznych modelov, preto je lepšie to spraviť len raz hneď na začiatku. (Zároveň to znamená, že na fitting predspracovania použijeme celú tréningovú množinu a nielen jej časť, ako by sa to stalo inak, keďže budeme používať krížovú validáciu.)

In [6]:
X_train_preproc = input_preproc.fit_transform(X_train)

Bayesovská optimalizácia

Ďalej si už prejdime časti priamo súvisiace s bayesovskou optimalizáciou. Čo potrebujeme urobiť v prvom rade je zadefinovať účelovú funkciu, ktorú má optimalizátor minimalizovať.

Keďže naším cieľom bude nájsť hyperparametre, pri ktorých náš model dosiahne najlepšie výsledky, vstupným argumentom budú tieto hyperparametre. S ich pomocou skonštruujeme model (rozhodovací strom na báze triedy DecisionTreeClassifier).

Správnosť modelu vyhodnotíme pomocou $k$-násobnej krížovej validácie. (Cross-validation. Tréningové dáta rozdelíme na $k$ častí, pričom zakaždým jednu časť použijeme na testovanie a zvyšné na trénovanie. Keď takto otestujeme model na všetkých kombináciách tréningových a testovacích množín, výslednú správnosť určíme ako priemer správností z jednotlivých behov.)

In [7]:
def objective(params):
    model = DecisionTreeClassifier(**params)
    
    score = cross_validate(model, X_train_preproc, Y_train,
                           scoring='f1_macro',
                           cv=10, n_jobs=10)['test_score'].mean()
    print("Score {:.3f} params {}".format(score, params))

    # znamienko mínus, pretože chceme čo najvyššiu správnosť,
    # ale hodnota účelovej funkcie sa bude minimalizovať
    return -score

Ďalej potrebujeme nakonfigurovať prehľadávaný priestor: t.j. špecifikovať, akými hyperparametrami disponuje naša metóda a určiť, aké hodnoty môžu nadobúdať. Začnime teda tým, že si zobrazíme dokumentáciu ku triede DecisionTreeClassifier:

In [8]:
print(DecisionTreeClassifier.__doc__)
A decision tree classifier.

    Read more in the :ref:`User Guide <tree>`.

    Parameters
    ----------
    criterion : string, optional (default="gini")
        The function to measure the quality of a split. Supported criteria are
        "gini" for the Gini impurity and "entropy" for the information gain.

    splitter : string, optional (default="best")
        The strategy used to choose the split at each node. Supported
        strategies are "best" to choose the best split and "random" to choose
        the best random split.

    max_depth : int or None, optional (default=None)
        The maximum depth of the tree. If None, then nodes are expanded until
        all leaves are pure or until all leaves contain less than
        min_samples_split samples.

    min_samples_split : int, float, optional (default=2)
        The minimum number of samples required to split an internal node:

        - If int, then consider `min_samples_split` as the minimum number.
        - If float, then `min_samples_split` is a fraction and
          `ceil(min_samples_split * n_samples)` are the minimum
          number of samples for each split.

        .. versionchanged:: 0.18
           Added float values for fractions.

    min_samples_leaf : int, float, optional (default=1)
        The minimum number of samples required to be at a leaf node.
        A split point at any depth will only be considered if it leaves at
        least ``min_samples_leaf`` training samples in each of the left and
        right branches.  This may have the effect of smoothing the model,
        especially in regression.

        - If int, then consider `min_samples_leaf` as the minimum number.
        - If float, then `min_samples_leaf` is a fraction and
          `ceil(min_samples_leaf * n_samples)` are the minimum
          number of samples for each node.

        .. versionchanged:: 0.18
           Added float values for fractions.

    min_weight_fraction_leaf : float, optional (default=0.)
        The minimum weighted fraction of the sum total of weights (of all
        the input samples) required to be at a leaf node. Samples have
        equal weight when sample_weight is not provided.

    max_features : int, float, string or None, optional (default=None)
        The number of features to consider when looking for the best split:

            - If int, then consider `max_features` features at each split.
            - If float, then `max_features` is a fraction and
              `int(max_features * n_features)` features are considered at each
              split.
            - If "auto", then `max_features=sqrt(n_features)`.
            - If "sqrt", then `max_features=sqrt(n_features)`.
            - If "log2", then `max_features=log2(n_features)`.
            - If None, then `max_features=n_features`.

        Note: the search for a split does not stop until at least one
        valid partition of the node samples is found, even if it requires to
        effectively inspect more than ``max_features`` features.

    random_state : int, RandomState instance or None, optional (default=None)
        If int, random_state is the seed used by the random number generator;
        If RandomState instance, random_state is the random number generator;
        If None, the random number generator is the RandomState instance used
        by `np.random`.

    max_leaf_nodes : int or None, optional (default=None)
        Grow a tree with ``max_leaf_nodes`` in best-first fashion.
        Best nodes are defined as relative reduction in impurity.
        If None then unlimited number of leaf nodes.

    min_impurity_decrease : float, optional (default=0.)
        A node will be split if this split induces a decrease of the impurity
        greater than or equal to this value.

        The weighted impurity decrease equation is the following::

            N_t / N * (impurity - N_t_R / N_t * right_impurity
                                - N_t_L / N_t * left_impurity)

        where ``N`` is the total number of samples, ``N_t`` is the number of
        samples at the current node, ``N_t_L`` is the number of samples in the
        left child, and ``N_t_R`` is the number of samples in the right child.

        ``N``, ``N_t``, ``N_t_R`` and ``N_t_L`` all refer to the weighted sum,
        if ``sample_weight`` is passed.

        .. versionadded:: 0.19

    min_impurity_split : float, (default=1e-7)
        Threshold for early stopping in tree growth. A node will split
        if its impurity is above the threshold, otherwise it is a leaf.

        .. deprecated:: 0.19
           ``min_impurity_split`` has been deprecated in favor of
           ``min_impurity_decrease`` in 0.19. The default value of
           ``min_impurity_split`` will change from 1e-7 to 0 in 0.23 and it
           will be removed in 0.25. Use ``min_impurity_decrease`` instead.

    class_weight : dict, list of dicts, "balanced" or None, default=None
        Weights associated with classes in the form ``{class_label: weight}``.
        If not given, all classes are supposed to have weight one. For
        multi-output problems, a list of dicts can be provided in the same
        order as the columns of y.

        Note that for multioutput (including multilabel) weights should be
        defined for each class of every column in its own dict. For example,
        for four-class multilabel classification weights should be
        [{0: 1, 1: 1}, {0: 1, 1: 5}, {0: 1, 1: 1}, {0: 1, 1: 1}] instead of
        [{1:1}, {2:5}, {3:1}, {4:1}].

        The "balanced" mode uses the values of y to automatically adjust
        weights inversely proportional to class frequencies in the input data
        as ``n_samples / (n_classes * np.bincount(y))``

        For multi-output, the weights of each column of y will be multiplied.

        Note that these weights will be multiplied with sample_weight (passed
        through the fit method) if sample_weight is specified.

    presort : bool, optional (default=False)
        Whether to presort the data to speed up the finding of best splits in
        fitting. For the default settings of a decision tree on large
        datasets, setting this to true may slow down the training process.
        When using either a smaller dataset or a restricted depth, this may
        speed up the training.

    Attributes
    ----------
    classes_ : array of shape = [n_classes] or a list of such arrays
        The classes labels (single output problem),
        or a list of arrays of class labels (multi-output problem).

    feature_importances_ : array of shape = [n_features]
        The feature importances. The higher, the more important the
        feature. The importance of a feature is computed as the (normalized)
        total reduction of the criterion brought by that feature.  It is also
        known as the Gini importance [4]_.

    max_features_ : int,
        The inferred value of max_features.

    n_classes_ : int or list
        The number of classes (for single output problems),
        or a list containing the number of classes for each
        output (for multi-output problems).

    n_features_ : int
        The number of features when ``fit`` is performed.

    n_outputs_ : int
        The number of outputs when ``fit`` is performed.

    tree_ : Tree object
        The underlying Tree object. Please refer to
        ``help(sklearn.tree._tree.Tree)`` for attributes of Tree object and
        :ref:`sphx_glr_auto_examples_tree_plot_unveil_tree_structure.py`
        for basic usage of these attributes.

    Notes
    -----
    The default values for the parameters controlling the size of the trees
    (e.g. ``max_depth``, ``min_samples_leaf``, etc.) lead to fully grown and
    unpruned trees which can potentially be very large on some data sets. To
    reduce memory consumption, the complexity and size of the trees should be
    controlled by setting those parameter values.

    The features are always randomly permuted at each split. Therefore,
    the best found split may vary, even with the same training data and
    ``max_features=n_features``, if the improvement of the criterion is
    identical for several splits enumerated during the search of the best
    split. To obtain a deterministic behaviour during fitting,
    ``random_state`` has to be fixed.

    See also
    --------
    DecisionTreeRegressor

    References
    ----------

    .. [1] https://en.wikipedia.org/wiki/Decision_tree_learning

    .. [2] L. Breiman, J. Friedman, R. Olshen, and C. Stone, "Classification
           and Regression Trees", Wadsworth, Belmont, CA, 1984.

    .. [3] T. Hastie, R. Tibshirani and J. Friedman. "Elements of Statistical
           Learning", Springer, 2009.

    .. [4] L. Breiman, and A. Cutler, "Random Forests",
           https://www.stat.berkeley.edu/~breiman/RandomForests/cc_home.htm

    Examples
    --------
    >>> from sklearn.datasets import load_iris
    >>> from sklearn.model_selection import cross_val_score
    >>> from sklearn.tree import DecisionTreeClassifier
    >>> clf = DecisionTreeClassifier(random_state=0)
    >>> iris = load_iris()
    >>> cross_val_score(clf, iris.data, iris.target, cv=10)
    ...                             # doctest: +SKIP
    ...
    array([ 1.     ,  0.93...,  0.86...,  0.93...,  0.93...,
            0.93...,  0.93...,  1.     ,  0.93...,  1.      ])
    

Úloha 1: Konfigurácia prehľadávaného priestoru

V nasledujúcej bunke definujte prehľadávaný priestor space pre hyperparametre rozhodovacieho stromu.


Priestor sa definuje slovníkom v nasledujúcom tvare:

space = {
    # kategorická premenná:
    'cat_var': hp.choice("cat_var", ["opt1", "opt2", "opt3"]),

    # celočíselná premenná z rovnomerného rozdelenia:
    'int_var': scope.int(hp.quniform("int_var", 1, 15, 1)),

    # reálnočíselná premenná z rovnomerného rozdelenia:
    'float_var': hp.uniform('float_var', 0.2, 1.0),
}

Ďalšie možnosti a podrobnejšia dokumentácia ku definovaniu parametrických priestorov sa dajú nájsť na wiki balíčka hyperopt.

In [0]:

Spustenie optimalizácie

Ďalej už môžeme spustiť samotnú optimalizáciu. Špecifikujeme pritom účelovú funkciu, prehľadávaný priestor, maximálny počet vyhodnotení účelovej funkcie a algoritmus. My používame algoritmus tpe, tzv. Tree-structured Parzen Estimator.

In [0]:
best = fmin(fn=objective,
            space=space,
            algo=tpe.suggest,
            max_evals=100
        )

Funkcia fmin navráti najlepšie nájdené riešenie. Následne ho dekódujeme pomocou funkcie space_eval, čím získame reprezentáciu, ktorú je už možné priamo použiť pri vytváraní nášho modelu.

In [0]:
best_params = space_eval(space, best)

Tréning modelu s najlepšími parametrami

Keď sme identifikovali najlepšie parametre, použijeme ich teraz, aby sme natrénovali nový model: tento raz už s použitím celej tréningovej množiny:

In [0]:
model = make_pipeline(
    input_preproc,
    DecisionTreeClassifier(**best_params)
)

model = model.fit(X_train, Y_train)

Testovanie

Na záver si model otestujeme na testovacích dátach. Zobrazíme si maticu zámen a klasické metriky. Úspešnosť predikcie by mala byť lepšia než pri predvolených hyperparametroch, ktoré sme používali v predchádzajúcom notebook-u.

In [0]:
y_test = model.predict(X_test)
In [0]:
cm = pd.crosstab(Y_test.values.reshape(-1), y_test,
                 rownames=['actual'],
                 colnames=['predicted'])
print(cm)
predicted    0   1
actual            
0          122  15
1           25  61
In [0]:
print("Accuracy = {}".format(accuracy_score(Y_test, y_test)))
print("Precision = {}".format(precision_score(Y_test, y_test)))
print("Recall = {}".format(recall_score(Y_test, y_test)))
Accuracy = 0.820627802690583
Precision = 0.8026315789473685
Recall = 0.7093023255813954

Optimalizácia hyperparametrov pre XGBoost

Vyskúšajte teraz celý postup zopakovať s inou klasifikačnou metódou: s metódou XGBClassifier z balíčka xgboost. Predefinovať bude potrebné najmä metódu objective, aby používala nový model a prehľadávaný priestor space tak aby zodpovedal hyperparametrom novej metódy.


In [0]:
from xgboost import XGBClassifier
In [0]: