def scores(conf):
parameters = {
'nn__n_iter': range(10, 25),
'nn__layers': [
# [Layer(type="Softmax", name="output")],
[Layer(type="Sigmoid", units=10, name="h0"), Layer(type="Softmax", name="output")],
[Layer(type="Sigmoid", units=20, name="h0"), Layer(type="Softmax", name="output")],
[Layer(type="Sigmoid", units=35, name="h0"), Layer(type="Softmax", name="output")],
[Layer(type="Sigmoid", units=50, name="h0"), Layer(type="Softmax", name="output")],
[Layer(type="Sigmoid", units=75, name="h0"), Layer(type="Softmax", name="output")],
# [Layer(type="Sigmoid", units=50, name="h0"), Layer(type="Sigmoid", units=50, name="h1"),
# Layer(type="Softmax", name="output")]
],
'nn__learning_rate': [.001, .01, .05],
# 'nn__learning_momentum': [.8, .9],
'nn__batch_size': [1, 10, 25, 50],
'nn__dropout_rate': [0, .1, .25, .5]
}
training_data, training_targets = learning_utils.getData(conf['datasets'], type=conf['type'], split=True,
balanced=conf['balanced'], shuffle=True)
clf = getNet(valid_size=0)
pipeline = learning_utils.getPipeline(training_data, clf, 'nn')
grid_search = GridSearchCV(pipeline, parameters, verbose=10)
return grid_search, training_data, training_targets
def train(conf):
training_data, training_targets, ids = learning_utils.getData(
conf["datasets"],
ratio=conf["ratio"],
type=conf["type"],
split=True,
balanced=conf["balanced"],
shuffle=True,
return_ids=True,
)
clf = svm.SVC(C=1, kernel="rbf")
clf1 = svm.SVC(C=100, kernel="rbf") # loss=conf['loss_type'])
# clf1 = svm.LinearSVC(C=.9)
clf.ids = ids
pipeline = learning_utils.getPipeline(training_data, clf, "svm")
pipeline1 = learning_utils.getPipeline(training_data, clf1, "svm")
scores = cross_validation.cross_val_score(pipeline, training_data, training_targets, cv=5, verbose=True, n_jobs=-1)
scores1 = cross_validation.cross_val_score(
pipeline1, training_data, training_targets, cv=5, verbose=True, n_jobs=-1
)
print scores
print scores1
return pipeline.fit(training_data, training_targets), training_data.columns.values
def train(conf):
training_data, training_targets = learning_utils.getData(conf['datasets'], type=conf['type'], split=True,
balanced=conf['balanced'], shuffle=True)
clf = tree.DecisionTreeClassifier(criterion=conf['criterion'], splitter='best', max_depth=12)
# clf = RandomForestClassifier(criterion=conf['criterion'], max_depth=12)
pipeline = learning_utils.getPipeline(training_data, clf, 'decision_tree')
scores = cross_validation.cross_val_score(pipeline, training_data, training_targets, cv=5)
print("Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
return pipeline.fit(training_data, training_targets), training_data.columns.values
def scores(conf):
parameters = {
'svm__C': np.logspace(-2, 5, 8),
'svm__gamma': np.logspace(-9, 2, 12),
'svm__kernel': ['linear', 'rbf']
}
training_data, training_targets = learning_utils.getData(conf['datasets'], type=conf['type'], split=True,
balanced=conf['balanced'], shuffle=True)
clf = svm.SVC()
pipeline = learning_utils.getPipeline(training_data, clf, 'svm')
grid_search = GridSearchCV(pipeline, parameters, n_jobs=-1, verbose=1)
learning_utils.gs(grid_search, training_data, training_targets)
def scores(conf):
parameters = {
"svm__C": np.logspace(-2, 5, 8),
"svm__gamma": np.logspace(-9, 2, 12),
"svm__kernel": ["linear", "rbf"],
}
training_data, training_targets = learning_utils.getData(
conf["datasets"], type=conf["type"], split=True, balanced=conf["balanced"], shuffle=True
)
clf = svm.SVC()
pipeline = learning_utils.getPipeline(training_data, clf, "svm")
grid_search = GridSearchCV(pipeline, parameters, n_jobs=-1, verbose=1)
learning_utils.gs(grid_search, training_data, training_targets)
def train(conf):
training_data, training_targets, ids = learning_utils.getData(conf['datasets'], ratio=conf['ratio'],
type=conf['type'], split=True,
balanced=conf['balanced'], shuffle=True,
return_ids=True)
clf = svm.SVC(C=1, kernel='rbf')
clf1 = svm.SVC(C=100, kernel='rbf') # loss=conf['loss_type'])
# clf1 = svm.LinearSVC(C=.9)
clf.ids = ids
pipeline = learning_utils.getPipeline(training_data, clf, 'svm')
pipeline1 = learning_utils.getPipeline(training_data, clf1, 'svm')
scores = cross_validation.cross_val_score(pipeline, training_data, training_targets, cv=5, verbose=True, n_jobs=-1)
scores1 = cross_validation.cross_val_score(pipeline1, training_data, training_targets, cv=5, verbose=True,
n_jobs=-1)
print scores
print scores1
return pipeline.fit(training_data, training_targets), training_data.columns.values
def getData(size, ratio, features, balanced, type, return_ids):
complete_data = learning_utils.getData(size, split=False, balanced=balanced, type=type, return_ids=return_ids)
if features is not None and len(features) > 0 and type != "md" and type != "mir":
features.append('peak_cat')
complete_data = complete_data[features]
threshold = int(complete_data.shape[0] * (1 - ratio))
if return_ids:
ids = complete_data['id']
complete_data.drop('id', axis=1, inplace=True)
else:
ids = None
training_data = complete_data[:threshold]
test_data = complete_data[threshold:]
training_targets = training_data['peak_cat']
training_data.drop('peak_cat', axis=1, inplace=True)
test_targets = test_data['peak_cat']
test_data.drop('peak_cat', axis=1, inplace=True)
return training_data, training_targets, test_data, test_targets, ids
def scores(conf):
if conf['tree'] == 'tree':
parameters = {
'tree__criterion': ['gini', 'entropy'],
'tree__splitter': ['best', 'random'],
'tree__max_features': [.1, .2, .5, 'sqrt', 'log2', None],
'tree__max_depth': range(1, 20),
'tree__presort': [True, False]
}
clf = tree.DecisionTreeClassifier()
elif conf['tree'] == 'random':
parameters = {
'tree__criterion': ['gini', 'entropy'],
'tree__n_estimators': range(5, 50),
'tree__max_features': [.1, .2, .5, 'sqrt', 'log2', None],
'tree__max_depth': range(1, 20),
'tree__bootstrap': [True, False],
}
clf = ensemble.RandomForestClassifier()
elif conf['tree'] == 'extra':
parameters = {
'tree__criterion': ['gini', 'entropy'],
'tree__n_estimators': range(5, 50),
'tree__max_features': [.1, .2, .5, 'sqrt', 'log2', None],
'tree__max_depth': range(1, 20),
'tree__bootstrap': [True, False],
}
clf = ensemble.ExtraTreesClassifier()
training_data, training_targets = learning_utils.getData(conf['datasets'], type=conf['type'], split=True,
balanced=conf['balanced'], shuffle=True)
pipeline = learning_utils.getPipeline(training_data, clf, 'tree')
grid_search = GridSearchCV(pipeline, parameters, n_jobs=-1, verbose=1)
learning_utils.gs(grid_search, training_data, training_targets)
