def mlp_titanic():
"""
constant_params = {
'random_state': 0,
'solver': 'lbfgs', # показал лучше всех
'early_stopping': False, # флаг возможности ранней остановки градиентного спуска
'learning_rate': 'invscaling',
'max_iter': 1000, # максимальное количество итераций градиентного спуска
}
MLP - model1: acc = 0.8731117824773413 %, tested 331 total.
MLP - model2: acc = 0.9063444108761329 %, tested 331 total.
MLP - model3: acc = 0.9274924471299094 %, tested 331 total.
"""
complete = False
titanic = Titanic('../data/titanic/')
train = titanic.load_train(complete)
test = titanic.load_test(complete)
constant_params = {
'random_state': 0,
'solver': 'lbfgs',
'early_stopping': False,
'learning_rate': 'invscaling',
'learning_rate_init': 0.001,
'max_iter': 1000,
'activation': 'relu'
}
model1(train, test, constant_params)
model2(train, test, constant_params)
model3(train, test, constant_params)
def main():
complete = True
titanic = Titanic('../data/titanic/')
train, test = titanic.load_train(complete), titanic.load_test(complete)
nb(train, test)
regressions(train, test)
def titanic():
titanic_train = Titanic('../../data/titanic/').load_train(complete=False)
X, y = titanic_train.data, titanic_train.target
survived = y == 1
died = y == 0
p = 35
X_emb = TSNE(n_components=2, perplexity=p, random_state=0,
verbose=1).fit_transform(X)
plt.figure()
plt.title("t-SNE for Titanic. perplexity={}, persons={}".format(p, len(y)))
plt.scatter(X_emb[survived, 0],
X_emb[survived, 1],
c='g',
marker='.',
label='Survived')
plt.scatter(X_emb[died, 0],
X_emb[died, 1],
c='r',
marker='x',
label='Died')
plt.legend(loc='best', shadow=False, scatterpoints=1)
plt.show()
def titanic_lda_pca():
complete = False
titanic = Titanic('../data/titanic/')
train = titanic.load_train(complete)
test = titanic.load_test(complete)
X, y = train.data, train.target
# todo join train + test -> PCA -> split again
train_pca = PCA(n_components=2, random_state=0)
train_pca_X = train_pca.fit_transform(X)
test_pca = PCA(n_components=2, random_state=0)
test_pca_X = test_pca.fit_transform(test.data)
m = RandomForestClassifier(random_state=0, max_depth=4, min_samples_split=25, min_samples_leaf=10)
m.fit(train_pca_X, y) # todo bokeh https://bokeh.pydata.org/en/latest/docs/gallery.html
survival_prediction = m.predict(test_pca_X)
# Статистика
print('Titanic PCA stats:')
print('explained variance ratio (first two components): {}'.format(train_pca.explained_variance_ratio_))
print('Components_: \n', pd.DataFrame(train_pca.components_, ['PC1', 'PC2']))
print('RF for PCA data: acc = {}%, tested {} total.'.format((survival_prediction == test.target).mean(),
len(survival_prediction)))
train_lda = LinearDiscriminantAnalysis(n_components=2)
traind_lda_X = train_lda.fit_transform(X, y)
test_lda = LinearDiscriminantAnalysis(n_components=2)
test_lda_X = test_lda.fit_transform(test.data, test.target)
m = RandomForestClassifier(random_state=0, max_depth=4, min_samples_split=25, min_samples_leaf=10)
m.fit(traind_lda_X, y)
survival_prediction = m.predict(test_lda_X)
print('RF for LDA data: acc = {}%, tested {} total.'.format((survival_prediction == test.target).mean(),
len(survival_prediction)))
pca_viz(train_pca_X, y)
lda_viz(traind_lda_X, y)
plt.show()
def main():
"""
https://www.kaggle.com/c/titanic/data
"""
complete = False
titanic = Titanic('../data/titanic/')
train, test = titanic.load_train(complete), titanic.load_test(complete)
params = [
# 1
{},
# 2
{
"max_depth": 10,
"min_samples_split": 10,
"min_samples_leaf": 50,
"max_features": 2
},
# 3
{
"max_depth": 4,
"min_samples_split": 10,
"min_samples_leaf": 50,
"max_features": 3
},
# 4
{
"max_depth": 10,
"min_samples_split": 40,
"min_samples_leaf": 10,
"max_features": 2
},
# 5
{
"max_depth": 10,
"min_samples_split": 2,
"min_samples_leaf": 1,
"max_features": 3
},
# 6
{
"max_depth": 10,
"min_samples_split": 25,
"min_samples_leaf": 10,
"max_features": None
},
# 7
{
"max_depth": 10,
"min_samples_split": 25,
"min_samples_leaf": 10,
"max_features": 6
},
]
for i, p in enumerate(params):
dt = DecisionTreeClassifier(random_state=5, **p)
dt.fit(train.data, train.target)
survival_prediction = dt.predict(test.data)
print('DT {}: acc = {}%, tested {} total.'.format(
i + 1,
np.round((survival_prediction == test.target).mean(), 4) * 100,
len(survival_prediction)))
dt_viz(dt, train.feature_names, train.target_names,
"dt_titanic_{}_model.tmp".format(i + 1))
def main():
titanic = Titanic('../data/titanic/')
train, test = titanic.load_train(False), titanic.load_test(False)
params = [
#1
{},
#2
{
"kernel": "rbf",
"C": 1,
"max_iter": -1,
"tol": 1e-3
},
#3
{
"kernel": "rbf",
"C": 1,
"max_iter": -1,
"tol": 1e-5
},
#4
{
"kernel": "rbf",
"C": 3,
"max_iter": -1,
"tol": 1e-3
},
#5
{
"kernel": "rbf",
"C": 3,
"max_iter": -1,
"tol": 1e-5
},
#6
{
"kernel": "linear",
"C": 1,
"max_iter": 10 * 10**6,
"tol": 1e-3
},
#7
{
"kernel": "linear",
"C": 1,
"max_iter": 10 * 10**6,
"tol": 1e-5
},
#8
{
"kernel": "linear",
"C": 3,
"max_iter": 10 * 10**6,
"tol": 1e-3
},
#9
{
"kernel": "linear",
"C": 3,
"max_iter": 10 * 10**6,
"tol": 1e-5
}
]
params = [
{
"kernel": "linear",
"C": 1,
"max_iter": 10 * 10**1,
"tol": 0.001
},
]
for i, p in enumerate(params):
m = SVC(random_state=0, **p)
m.fit(train.data, train.target)
survival_prediction = m.predict(test.data)
print('SVM {}: acc = {}%, tested {} total.'.format(
i + 1,
np.round((survival_prediction == test.target).mean(), 4) * 100,
len(survival_prediction)))
def main():
params = [{}, {
"max_depth": 15,
"n_estimators": 13,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 2,
}, {
"max_depth": 15,
"n_estimators": 5,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 2
}, {
"max_depth": 15,
"n_estimators": 2,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 2
}, {
"max_depth": 15,
"n_estimators": 13,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 3
}, {
"max_depth": 15,
"n_estimators": 5,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 3
}, {
"max_depth": 15,
"n_estimators": 2,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 3
}, {
"max_depth": 15,
"n_estimators": 13,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 7
}, {
"max_depth": 15,
"n_estimators": 5,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 7
}, {
"max_depth": 15,
"n_estimators": 2,
"min_samples_split": 50,
"min_samples_leaf": 10,
"max_features": 7
}]
complete = True
titanic = Titanic('../data/titanic/')
train, test = titanic.load_train(complete), titanic.load_test(complete)
for i, p in enumerate(params):
m = RandomForestClassifier(random_state=0, **p)
m.fit(train.data, train.target)
survival_prediction = m.predict(test.data)
print('Random Forest: acc = {}%, tested {} total.'.format(
(survival_prediction == test.target).mean(),
len(survival_prediction)))