def load_data_cache(use_test=False, out_ohe=True):
test_size = 0.33
valid_size = 0.33 * (1 - test_size)
if use_test:
print("!!! USING TEST DATA !!!")
(
(X_train, y_train),
(X_valid, y_valid),
(X_test, y_test),
categorical_indicator,
) = albert.load_data(
random_state=RANDOM_STATE,
test_size=test_size,
valid_size=valid_size,
categoricals_to_integers=True,
)
X_train = np.concatenate([X_train, X_valid])
y_train = np.concatenate([y_train, y_valid])
X_valid, y_valid = X_test, y_test
else:
(
(X_train, y_train),
(X_valid, y_valid),
_,
categorical_indicator,
) = albert.load_data(
random_state=RANDOM_STATE,
test_size=test_size,
valid_size=valid_size,
categoricals_to_integers=True,
)
# Replace missing values with mean value
# https://scikit-learn.org/stable/modules/impute.html
print("Replacing missing values")
imp_mean = SimpleImputer(missing_values=np.nan, strategy="mean")
X_train = imp_mean.fit_transform(X_train)
X_valid = imp_mean.transform(X_valid)
# Min Max => Std scaler preprocessing for non categorical variables
for i, (categorical, _) in enumerate(categorical_indicator):
if not categorical:
scaler = minmaxstdscaler()
X_train[:, i:i + 1] = scaler.fit_transform(X_train[:, i:i + 1])
X_valid[:, i:i + 1] = scaler.transform(X_valid[:, i:i + 1])
# One Hot Encoding of Outputs
if out_ohe:
prepro_output = preprocessing.OneHotEncoder()
y_train = y_train.reshape(-1, 1)
y_valid = y_valid.reshape(-1, 1)
y_train = prepro_output.fit_transform(y_train).toarray()
y_valid = prepro_output.transform(y_valid).toarray()
print(f"X_train shape: {np.shape(X_train)}")
print(f"y_train shape: {np.shape(y_train)}")
print(f"X_valid shape: {np.shape(X_valid)}")
print(f"y_valid shape: {np.shape(y_valid)}")
return (X_train, y_train), (X_valid, y_valid), categorical_indicator
def load_data_cache(use_test=False):
# Random state
random_state = np.random.RandomState(seed=42)
if use_test:
print("!!! USING TEST DATA !!!")
(X_train, y_train), (X_valid,
y_valid), (X_test, y_test) = airlines.load_data(
random_state=random_state,
test_size=0.33,
valid_size=0.33 * (1 - 0.33))
X_train = np.concatenate([X_train, X_valid])
y_train = np.concatenate([y_train, y_valid])
X_valid, y_valid = X_test, y_test
else:
(X_train, y_train), (X_valid, y_valid), _ = airlines.load_data(
random_state=random_state,
test_size=0.33,
valid_size=0.33 * (1 - 0.33))
prepro_output = preprocessing.OneHotEncoder()
y_train = y_train.reshape(-1, 1)
y_valid = y_valid.reshape(-1, 1)
y_train = prepro_output.fit_transform(y_train).toarray()
y_valid = prepro_output.transform(y_valid).toarray()
prepro_input = minmaxstdscaler()
X_train = prepro_input.fit_transform(X_train)
X_valid = prepro_input.transform(X_valid)
print(f"X_train shape: {np.shape(X_train)}")
print(f"y_train shape: {np.shape(y_train)}")
print(f"X_valid shape: {np.shape(X_valid)}")
print(f"y_valid shape: {np.shape(y_valid)}")
return (X_train, y_train), (X_valid, y_valid)
def __init__(
self,
clf=KNeighborsClassifier(n_jobs=4),
load_data_func=lambda: load_breast_cancer(return_X_y=True),
preproc=minmaxstdscaler(),
seed=42,
):
super().__init__(
clf=clf, load_data_func=load_data_func, preproc=preproc, seed=seed
)
def run(config: dict, load_data: callable) -> float:
"""Run function which can be used for AutoML classification.
Args:
config (dict): [description]
load_data (callable): [description]
Returns:
float: [description]
"""
seed = 42
config["random_state"] = seed
X, y = load_data()
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.33,
random_state=seed)
preproc = minmaxstdscaler()
X_train = preproc.fit_transform(X_train)
X_test = preproc.transform(X_test)
mapping = CLASSIFIERS
clf_class = mapping[config["classifier"]]
# keep parameters possible for the current classifier
sig = signature(clf_class)
clf_allowed_params = list(sig.parameters.keys())
clf_params = {
k: v
for k, v in config.items()
if k in clf_allowed_params and not (v in ["nan", "NA"])
}
if "n_jobs" in clf_allowed_params: # performance parameter
clf_params["n_jobs"] = 8
try: # good practice to manage the fail value yourself...
clf = clf_class(**clf_params)
clf.fit(X_train, y_train)
fit_is_complete = True
except:
fit_is_complete = False
if fit_is_complete:
y_pred = clf.predict(X_test)
acc = accuracy_score(y_test, y_pred)
else:
acc = -1.0
return acc
def __init__(
self,
clf=RandomForestRegressor(n_jobs=4, random_state=42),
load_data_func=lambda: load_boston(return_X_y=True),
preproc=minmaxstdscaler(),
seed=42,
):
super().__init__(clf=clf,
load_data_func=load_data_func,
preproc=preproc,
seed=seed)
def __init__(
self,
clf=None,
load_data_func=lambda: load_breast_cancer(return_X_y=True),
preproc=minmaxstdscaler(),
seed=42,
):
self.clf = clf
self.seed = seed
self.load_data_func = load_data_func
self.preproc = preproc
def load_data_cache():
# Random state
random_state = np.random.RandomState(seed=42)
(X_train, y_train), (X_valid, y_valid), _ = covertype.load_data(
random_state=random_state
)
prepro_output = preprocessing.OneHotEncoder()
y_train = y_train.reshape(-1, 1)
y_valid = y_valid.reshape(-1, 1)
y_train = prepro_output.fit_transform(y_train).toarray()
y_valid = prepro_output.transform(y_valid).toarray()
prepro_input = minmaxstdscaler()
X_train = prepro_input.fit_transform(X_train)
X_valid = prepro_input.transform(X_valid)
print(f"X_train shape: {np.shape(X_train)}")
print(f"y_train shape: {np.shape(y_train)}")
print(f"X_valid shape: {np.shape(X_valid)}")
print(f"y_valid shape: {np.shape(y_valid)}")
return (X_train, y_train), (X_valid, y_valid)
def run_autosklearn1(config: dict, load_data: callable) -> float:
"""Run function which can be used for AutoML classification.
It has to be used with the ``deephyper.sklearn.classifier.problem_autosklearn1`` problem definition which corresponds to:
.. code-block::
Configuration space object:
Hyperparameters:
C, Type: UniformFloat, Range: [1e-05, 10.0], Default: 0.01, on log-scale
alpha, Type: UniformFloat, Range: [1e-05, 10.0], Default: 0.01, on log-scale
classifier, Type: Categorical, Choices: {RandomForest, Logistic, AdaBoost, KNeighbors, MLP, SVC, XGBoost}, Default: RandomForest
gamma, Type: UniformFloat, Range: [1e-05, 10.0], Default: 0.01, on log-scale
kernel, Type: Categorical, Choices: {linear, poly, rbf, sigmoid}, Default: linear
max_depth, Type: UniformInteger, Range: [2, 100], Default: 14, on log-scale
n_estimators, Type: UniformInteger, Range: [1, 2000], Default: 45, on log-scale
n_neighbors, Type: UniformInteger, Range: [1, 100], Default: 50
Conditions:
(C | classifier == 'Logistic' || C | classifier == 'SVC')
(gamma | kernel == 'rbf' || gamma | kernel == 'poly' || gamma | kernel == 'sigmoid')
(n_estimators | classifier == 'RandomForest' || n_estimators | classifier == 'AdaBoost')
alpha | classifier == 'MLP'
kernel | classifier == 'SVC'
max_depth | classifier == 'RandomForest'
n_neighbors | classifier == 'KNeighbors'
Args:
config (dict): an hyperparameter configuration ``dict`` corresponding to the ``deephyper.sklearn.classifier.problem_autosklearn1``.
load_data (callable): a function returning data as Numpy arrays ``(X, y)``.
Returns:
float: returns the accuracy on the validation set.
"""
seed = 42
config["random_state"] = seed
X, y = load_data()
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.33,
random_state=seed)
preproc = minmaxstdscaler()
X_train = preproc.fit_transform(X_train)
X_test = preproc.transform(X_test)
mapping = CLASSIFIERS
clf_class = mapping[config["classifier"]]
# keep parameters possible for the current classifier
sig = signature(clf_class)
clf_allowed_params = list(sig.parameters.keys())
clf_params = {
k: v
for k, v in config.items()
if k in clf_allowed_params and not (v in ["nan", "NA"])
}
if "n_jobs" in clf_allowed_params: # performance parameter
clf_params["n_jobs"] = 8
try: # good practice to manage the fail value yourself...
clf = clf_class(**clf_params)
clf.fit(X_train, y_train)
fit_is_complete = True
except:
fit_is_complete = False
if fit_is_complete:
y_pred = clf.predict(X_test)
acc = accuracy_score(y_test, y_pred)
else:
acc = -1.0
return acc