def rf():
"""
Submission: rf_0708_01.csv
3000 trees
E_val: 0.871837
E_in: 0.999998
E_out: 0.882316801296279
15000 trees
E_val: 0.872011
E_in: 0.999998
E_out: 0.8824869811781106
30000 trees
E_val: 0.871928
E_in:
E_out:
depth=4; 12000 trees
E_val: 0.969158
E_in:
E_out:
"""
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
import numpy as np
X, y = dataset.load_train(depth=1)
raw_scaler = StandardScaler()
raw_scaler.fit(np.r_[X, dataset.load_test()])
X_scaled = raw_scaler.transform(X)
del X
import gc
gc.collect()
rf = RandomForestClassifier(n_estimators=12000, oob_score=True, n_jobs=-1,
class_weight='auto')
rf.fit(X_scaled, y)
logger.debug('RandomForestClassifier fitted')
logger.debug('E_val(oob): %f', rf.oob_score_)
logger.debug('E_in(full): %f', Util.auc_score(rf, X_scaled, y))
X, y = dataset.load_train()
X_scaled = raw_scaler.transform(X)
logger.debug('E_in (depth=0): %f', Util.auc_score(rf, X_scaled, y))
del X
gc.collect()
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('rf', rf)]), 'rf_0708_01')
logger.debug('caching fitted RandomForestClassifier')
IO.cache(rf, Path.of_cache('rf.RandomForestClassifier.12000.pkl'))
logger.debug('cached fitted RandomForestClassifier')
def run(args: Namespace):
"""
actual function which is doing some task
Args:
args: program arguments
"""
Trainer(args, dataset.load_train(args.train)).run()
def svc_test2():
"""
Submission:
E_val:
E_in:
E_out:
"""
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC
from sklearn.cross_validation import StratifiedKFold
from sklearn.calibration import CalibratedClassifierCV
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
svc = SVC(kernel='linear', class_weight='auto', cache_size=10240)
svc.fit(X_scaled, y)
isotonic = CalibratedClassifierCV(svc, cv=StratifiedKFold(y, 5),
method='isotonic')
isotonic.fit(X_scaled, y)
logger.debug('Got best isotonic CalibratedClassifier.')
logger.debug('E_in (isotonic): %f', Util.auc_score(isotonic, X_scaled, y))
def q20(): def cross_validate(X, y, lamda): Xs = X[:40], X[40:80], X[80:120], X[120:160], X[160:] ys = y[:40], y[40:80], y[80:120], y[120:160], y[160:] reg = ridge.RidgeRegression(lamda) e_cv = 0.0 for i in range(5): X_val, y_val = Xs[i], ys[i] X_train = np.concatenate([Xs[j] for j in range(5) if i != j]) y_train = np.concatenate([ys[j] for j in range(5) if i != j]) reg.fit(X_train, y_train) e_cv += reg.evaluate(X_val, y_val, sign) return e_cv / 5 X_train, y_train = load_train() X_test, y_test = load_test() lamdas = np.logspace(-10, 2, 13) best = (1.0, None) for lamda in lamdas: e_cv = cross_validate(X_train, y_train, lamda) if e_cv <= best[0]: best = (e_cv, lamda) best_reg = ridge.RidgeRegression(best[1]) best_reg.fit(X_train, y_train) print "lamda: %e, E_in: %.3f, E_out: %.3f" % (best[1], best_reg.evaluate(X_train, y_train, sign), best_reg.evaluate(X_test, y_test, sign))
def bagging_lr():
"""
Submission: bagging_lr_0707_02.csv
E_val:
E_in:
E_out:
"""
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import BaggingClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
bag = BaggingClassifier(LogisticRegression(class_weight='auto'),
n_estimators=3000, oob_score=True, n_jobs=-1,
verbose=2)
logger.debug('E_val (oob): %f', bag.oob_score_)
logger.debug('E_in: %f', Util.auc_score(bag, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('bag', bag)]), 'bagging_lr_0707_02')
def lr_with_scale3():
"""
Check the performance of normalizing TEST SET.
Submission: lr_with_scale3_0707_04.csv
E_val:
E_in: 0.879233
E_out: 0.8770121701777971
Submission: lr_with_scale3_0712_01.csv
E_val:
E_in:
E_out:
"""
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import StandardScaler
from sklearn.cross_validation import cross_val_score
from sklearn.pipeline import Pipeline
import numpy as np
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(np.r_[X, dataset.load_test()])
X_scaled = raw_scaler.transform(X)
clf = LogisticRegression(C=0.03, class_weight='auto')
clf.fit(X_scaled, y)
logger.debug('E_in: %f', Util.auc_score(clf, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('lr', clf)]), 'lr_with_scale3_0712_01')
scores = cross_val_score(clf, X_scaled, y, scoring='roc_auc', n_jobs=-1)
logger.debug('E_val: %f <- %s', np.average(scores), scores)
def trainSVM(train_path,
classes,
image_size=32,
model_file='model.pkl',
download_flag=False):
print('Training with HOG feature and SVM')
if download_flag:
print('Downloading images')
for i in classes:
iurl.downloadImage(i + '.txt', train_path + '/' + i)
X, Y, ids, cls, testX, testY, ids, cls, num = dataset.load_train(
train_path, image_size, classes, 5, True)
print("Number of training data: {}".format(num))
rand = np.random.RandomState(321)
shuffle = rand.permutation(len(X))
# print(X)
# print(testX)
# print(Y)
# print(len(X[:-1]),len(Y))
X = X[shuffle]
Y = Y[shuffle]
# print(X)
# print(Y)
# pause()
clf = svm.SVC()
clf.fit(X, Y)
result = clf.predict(testX)
mask = result == testY
correct = np.count_nonzero(mask)
# print result
print("Accuracy: {}", correct * 100.0 / len(result))
joblib.dump(clf, model_file)
def svc_appr():
"""
Best params: {'C': 0.022139881953014046}
Submission:
E_val:
E_in:
E_out:
"""
from sklearn.svm import LinearSVC
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.cross_validation import StratifiedKFold
from sklearn.grid_search import RandomizedSearchCV
from scipy.stats import expon
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
svc = LinearSVC(dual=False, class_weight='auto')
rs = RandomizedSearchCV(svc, n_iter=50, scoring='roc_auc', n_jobs=-1,
cv=StratifiedKFold(y, 5), verbose=2,
param_distributions={'C': expon()})
rs.fit(X_scaled, y)
logger.debug('Got best SVC.')
logger.debug('Best params: %s', rs.best_params_)
logger.debug('Grid scores:')
for i, grid_score in enumerate(rs.grid_scores_):
print('\t%s' % grid_score)
logger.debug('Best score (E_val): %s', rs.best_score_)
logger.debug('E_in: %f', Util.auc_score(rs, X_scaled, y))
def __init__(self, train_path, image_size, classes, batch_size):
self.graph = tf.Graph()
with self.graph.as_default():
self.num_classes = len(classes)
self.image_size = image_size
images, labels, _, _ = load_train(train_path, image_size, classes)
self.X, self.labels = get_data_batch(images,
labels,
batch_size,
num_threads=8)
self.Y = tf.one_hot(self.labels,
depth=self.num_classes,
axis=1,
dtype=tf.float32)
self.batch_size = batch_size
# self.X, self.labels = get_batch_data('mnist', batch_size, 4)
# self.Y = tf.one_hot(self.labels, depth=10, axis=1, dtype=tf.float32)
self.build_arch()
self.loss()
self.model_summary()
self.global_step = tf.Variable(0,
name='global_step',
trainable=False)
self.optimizer = tf.train.AdamOptimizer()
self.train_op = self.optimizer.minimize(
self.total_loss, global_step=self.global_step)
def ada_boost_dt():
"""
Submission: ada_boost_dt_0707_03.csv
E_val: 0.854350
E_in: 0.889561
E_out: 0.8832315976033993
"""
from sklearn.ensemble import AdaBoostClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.cross_validation import cross_val_score
from sklearn.pipeline import Pipeline
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
ab = AdaBoostClassifier(n_estimators=300)
scores = cross_val_score(ab, X_scaled, y, cv=5, n_jobs=-1)
logger.debug('CV: %s', scores)
logger.debug('E_val: %f', sum(scores) / len(scores))
ab.fit(X_scaled, y)
logger.debug('E_in: %f', Util.auc_score(ab, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('ab', ab)]), 'ada_boost_dt_0707_03')
def trainCNN(train_path, classes, image_size = 32, model_file = 'model.pkl', download_flag = False):
if download_flag:
for i in classes:
iurl.downloadImage(i+'.txt', train_path+'/'+i)
#iurl.downloadImage(catsUrlFile,train_path+'/cats')
#iurl.downloadImage(dogsUrlFile,train_path+'/dogs')
X, Y, ids, cls, testX, testY, ids, cls, num= dataset.load_train(train_path,image_size,classes,5)
#testX, testY, ids, cls= dataset.load_train(test_path,image_size,classes)
print('Number of test data:{}'.format(num))
X = X.reshape([-1, image_size, image_size, 1])
testX = testX.reshape([-1, image_size, image_size, 1])
# Building convolutional network
network = input_data(shape=[None, image_size, image_size, 1], name='input')
network = conv_2d(network, 16, 3, activation='relu', regularizer="L2")
network = max_pool_2d(network, 2)
network = local_response_normalization(network)
network = conv_2d(network, 32, 3, activation='relu', regularizer="L2")
network = max_pool_2d(network, 2)
network = local_response_normalization(network)
network = fully_connected(network, 32, activation='tanh')
network = dropout(network, 0.8)
network = fully_connected(network, 64, activation='tanh')
network = dropout(network, 0.8)
network = fully_connected(network, len(classes), activation='softmax')
network = regression(network, optimizer='adam', learning_rate=0.001, loss='categorical_crossentropy', name='target')
# Training
model = tflearn.DNN(network, tensorboard_verbose=0)
print("start")
model.fit({'input': X}, {'target': Y}, n_epoch=20, shuffle = True, batch_size = 100, validation_set=({'input': testX}, {'target': testY}), snapshot_step=100, show_metric=True, run_id='convnet_mnist')
print(model.predict(testX))
model.save(model_file)
def rf2():
"""
Submission: rf2_0704_04.csv
3000 trees
E_val: 0.871431
E_in: 0.999998
E_out:
30000 trees
E_val:
E_in:
E_out:
"""
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
rf = RandomForestClassifier(n_estimators=30000, oob_score=True, n_jobs=-1,
class_weight='auto', max_features='log2')
rf.fit(X_scaled, y)
logger.debug('Eval(oob): %f', rf.oob_score_)
logger.debug('Ein: %f', Util.auc_score(rf, X_scaled, y))
IO.cache(rf, Path.of_cache('rf.RandomForestClassifier.log2.pkl'))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('rf', rf)]), 'rf2_0704_04')
def lr_with_scale2():
"""
Submission: lr_with_scale2_0704_03.csv
E_val:
E_in: 0.878996
E_out: 0.8768131004917349
"""
from sklearn.linear_model import LogisticRegressionCV
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
clf = LogisticRegressionCV(Cs=50, cv=5, scoring='roc_auc', n_jobs=-1,
class_weight='auto')
clf.fit(X_scaled, y)
logger.debug('Best C: %f', clf.C_[0])
logger.debug('Cs: %s', clf.Cs_)
logger.debug('Grid scores: %f', clf.scores_)
logger.debug('Ein: %f', Util.auc_score(clf, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('lr', clf)]), 'lr_with_scale2_0704_03')
def gbdt():
"""
Submission: gbdt_0708_02.csv
n_estimators: 1000, learning_rate: 0.1, subsample: 0.5
E_val: 0.858235
E_in: 0.908622
E_out: 0.8873906795559863
n_estimators: 500, learning_rate: 0.1, subsample: 0.5
E_val: 0.870976
E_in: 0.899593
E_out: 0.88711101837711
n_estimators: 3000, learning_rate: 0.1, subsample: 0.5
E_val: 0.836049
E_in: 0.936056
E_out: 0.8833930861722906
depth=4; n_estimators: 1000, learning_rate: 0.1, subsample: 0.5
E_val: 0.947301
E_in: 0.983812 (on depth=4) // 0.85089646325496504 (on depth=0)
E_out: 0.8855316272153549
"""
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
import numpy as np
gb = GradientBoostingClassifier(n_estimators=1000, learning_rate=0.1,
subsample=0.5)
d = 0
X, y = dataset.load_train(depth=d)
raw_scaler = StandardScaler()
raw_scaler.fit(np.r_[X, dataset.load_test()])
X_scaled = raw_scaler.transform(X)
gb.fit(X_scaled, y)
IO.cache(gb, Path.of_cache('gbdt.GradientBoostingClassifier.d%d.pkl' % d))
IO.dump_submission(Pipeline([('scaler', raw_scaler), ('gbdt', gb)]),
'gbdt_0708_02.1000.d%d' % d)
logger.debug('E_in(full): %f', Util.auc_score(gb, X_scaled, y))
X, y = dataset.load_train()
X_scaled = raw_scaler.transform(X)
logger.debug('E_in(depth=0): %f', Util.auc_score(gb, X_scaled, y))
def test(): X_train, y_train = load_train() X_test, y_test = load_test() lamda = 0 reg = ridge.RidgeRegression(lamda) reg.fit(X_train, y_train) e_in = reg.evaluate(X_train, y_train, sign) e_out = reg.evaluate(X_test, y_test, sign) print "E_in: %.3f, E_out: %.3f" % (e_in, e_out)
def lr_with_fs():
"""
Submission: lr_with_fs_0703_01.csv
E_val:
E_in:
E_out:
"""
from sklearn.linear_model import LogisticRegressionCV, LogisticRegression
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.cross_validation import StratifiedKFold
from sklearn.feature_selection import RFECV
import pylab as pl
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
pkl_path = Path.of_cache('lr_with_fs.RFECV.pkl')
rfe = IO.fetch_cache(pkl_path)
if rfe is None:
rfe = RFECV(estimator=LogisticRegression(class_weight='auto'),
cv=StratifiedKFold(y, 5), scoring='roc_auc')
rfe.fit(X_scaled, y)
IO.cache(rfe, pkl_path)
print("Optimal number of features : %d" % rfe.n_features_)
# Plot number of features VS. cross-validation scores
pl.figure()
pl.xlabel("Number of features selected")
pl.ylabel("Cross validation score (AUC)")
pl.plot(range(1, len(rfe.grid_scores_) + 1), rfe.grid_scores_)
pl.savefig('lr_with_fs.refcv')
X_pruned = rfe.transform(X_scaled)
new_scaler = StandardScaler()
new_scaler.fit(X_pruned)
X_new = new_scaler.transform(X_pruned)
clf = LogisticRegressionCV(cv=10, scoring='roc_auc', n_jobs=-1)
clf.fit(X_new, y)
print('CV scores: %s' % clf.scores_)
print('Ein: %f' % Util.auc_score(clf, X_new, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('rfe', rfe),
('scale_new', new_scaler),
('lr', clf)]), 'lr_with_fs_0703_01')
def gbdt_search():
"""
Grid search for best n_estimators.
Best params: {'loss': 'deviance', 'n_estimators': 100}
Submission: gbdt_search_0707_01.csv
E_val: 0.883786743214
E_in: 0.887785
E_out: 0.8848760405053878
"""
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.grid_search import GridSearchCV
from sklearn.cross_validation import StratifiedKFold
from sklearn.pipeline import Pipeline
import numpy as np
X, y = dataset.load_train()
raw_scaler = StandardScaler()
X_scaled = raw_scaler.fit_transform(X)
param_grid = {
'loss': ['deviance', 'exponential'],
'n_estimators': np.arange(100, 1001, 100)
}
params = {'learning_rate': 0.1, 'subsample': 0.5}
gb = GradientBoostingClassifier(**params)
grid = GridSearchCV(gb, param_grid, scoring='roc_auc', n_jobs=-1,
cv=StratifiedKFold(y, 5), refit=True, verbose=1)
grid.fit(X_scaled, y)
logger.debug('Got best GBDT.')
logger.debug('Grid scores: ')
for i, grid_score in enumerate(grid.grid_scores_):
print('\t%d00: %s' % (i + 1, grid_score))
logger.debug('Best score (E_val): %s', grid.best_score_)
logger.debug('Best params: %s', grid.best_params_)
IO.cache(grid, Path.of_cache('gbdt_search.GridSearchCV.pkl'))
X_test = dataset.load_test()
raw_scaler.fit(np.r_[X, X_test])
X_scaled = raw_scaler.transform(X)
params.update(grid.best_params_)
clf = GradientBoostingClassifier(**params)
clf.fit(X_scaled, y)
logger.debug('E_in: %f', Util.auc_score(grid, X_scaled, y))
IO.dump_submission(Pipeline([('scaler', raw_scaler),
('gbdt', grid)]), 'gbdt_search_0707_01')
def q15(): X_train, y_train = load_train() X_test, y_test = load_test() lamdas = np.logspace(-10, 2, 13) best = (1.0, None) for lamda in lamdas: reg = ridge.RidgeRegression(lamda) reg.fit(X_train, y_train) e_out = reg.evaluate(X_test, y_test, sign) if e_out <= best[0]: best = (e_out, reg) best_reg = best[1] print "lamda: %e, E_in: %.3f, E_out: %.3f" % ( best_reg.lamda, best_reg.evaluate(X_train, y_train, sign), best_reg.evaluate(X_test, y_test, sign))
def gbdt_grid():
"""
Grid search for best params.
Best params: {'learning_rate': 0.05, 'subsample': 0.3}
Submission: gbdt_grid_0706_03.csv
E_val: 0.860118290628
E_in: 0.882949
E_out: 0.8809314555068068
"""
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.grid_search import GridSearchCV
from sklearn.cross_validation import StratifiedKFold
from sklearn.pipeline import Pipeline
import numpy as np
X, y = dataset.load_train()
raw_scaler = StandardScaler()
X_scaled = raw_scaler.fit_transform(X)
param_grid = {
'learning_rate': [0.05, 0.1],
'subsample': [0.3, 0.5, 0.7]
}
grid = GridSearchCV(GradientBoostingClassifier(n_estimators=3000),
param_grid, scoring='roc_auc', n_jobs=-1,
cv=StratifiedKFold(y, 5), refit=False, verbose=1)
grid.fit(X_scaled, y)
logger.debug('Got best GBDT.')
logger.debug('Grid scores: %s', grid.grid_scores_)
logger.debug('Best score (E_val): %s', grid.best_score_)
logger.debug('Best params: %s', grid.best_params_)
X_test = dataset.load_test()
raw_scaler.fit_transform(np.r_[X, X_test])
X_scaled = raw_scaler.transform(X)
clf = GradientBoostingClassifier(**grid.best_params_)
clf.fit(X_scaled, y)
IO.cache(grid, Path.of_cache('gbdt_grid.GridSearchCV.pkl'))
logger.debug('E_in: %f', Util.auc_score(clf, X_scaled, y))
IO.dump_submission(Pipeline([('scaler', raw_scaler),
('gbdt', clf)]), 'gbdt_grid_0706_03')
def svc():
"""
Submission: svc_0703_04.csv
E_val:
E_in:
E_out:
"""
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.svm import SVC
from sklearn.cross_validation import StratifiedKFold
from sklearn.grid_search import RandomizedSearchCV
from sklearn.calibration import CalibratedClassifierCV
from scipy.stats import expon
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
svc = SVC(kernel='linear', class_weight='auto')
rs = RandomizedSearchCV(svc, n_iter=50, scoring='roc_auc', n_jobs=-1,
cv=StratifiedKFold(y, 5),
param_distributions={'C': expon()})
rs.fit(X_scaled, y)
logger.debug('Got best SVC.')
logger.debug('Grid scores: %s', rs.grid_scores_)
logger.debug('Best score (E_val): %s', rs.best_score_)
logger.debug('Best params: %s', rs.best_params_)
IO.cache(rs, Path.of_cache('svc.RandomizedSearchCV.SVC.pkl'))
svc = rs.best_estimator_
IO.cache(rs, Path.of_cache('svc.SVC.pkl'))
isotonic = CalibratedClassifierCV(svc, cv=StratifiedKFold(y, 5),
method='isotonic')
isotonic.fit(X_scaled, y)
IO.cache(rs, Path.of_cache('svc.CalibratedClassifierCV.isotonic.pkl'))
logger.debug('Got best isotonic CalibratedClassifier.')
logger.debug('E_in (isotonic): %f', Util.auc_score(isotonic, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('svc', isotonic)]), 'svc_0703_04')
def q17(): X, y = load_train() X_train, X_val = X[:120], X[120:] y_train, y_val = y[:120], y[120:] X_test, y_test = load_test() lamdas = np.logspace(-10, 2, 13) best = (1.0, None) for lamda in lamdas: reg = ridge.RidgeRegression(lamda) reg.fit(X_train, y_train) e_val = reg.evaluate(X_val, y_val, sign) if e_val <= best[0]: best = (e_val, reg) best_reg = best[1] print "lamda: %e, E_train: %.3f, E_val: %.3f, E_out: %.3f" % ( best_reg.lamda, best_reg.evaluate(X_train, y_train, sign), best_reg.evaluate(X_val, y_val, sign), best_reg.evaluate(X_test, y_test, sign))
def erf():
"""
Submission: erf_0705_01.csv
3000 trees
E_val: 0.870800
E_in: 0.999998
E_out:
15000 trees
E_val:
E_in:
E_out:
"""
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.ensemble import ExtraTreesClassifier
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
del X
rf = ExtraTreesClassifier(n_estimators=3000, oob_score=True, n_jobs=-1,
class_weight='auto', bootstrap=True)
rf.fit(X_scaled, y)
logger.debug('ExtraTreesClassifier fitted')
import gc
gc.collect()
logger.debug('Eval(oob): %f', rf.oob_score_)
logger.debug('Ein: %f', Util.auc_score(rf, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('erf', rf)]), 'erf_0705_01')
logger.debug('caching fitted ExtraTreesClassifier')
IO.cache(rf, Path.of_cache('erf.ExtraTreesClassifier.auto.pkl'))
logger.debug('cached fitted ExtraTreesClassifier')
def lr_with_fs1():
"""
Submission: lr_with_fs1_0703_03.csv
E_val:
E_in: 0.876954
E_out:
"""
from sklearn.linear_model import LogisticRegressionCV, LogisticRegression
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
pkl_path = Path.of_cache('lr_with_fs1.LR.FS.pkl')
lr = IO.fetch_cache(pkl_path)
if lr is None:
lr = LogisticRegression(class_weight='auto')
lr.fit(X_scaled, y)
IO.cache(lr, pkl_path)
X_pruned = lr.transform(X_scaled)
new_scaler = StandardScaler()
new_scaler.fit(X_pruned)
X_new = new_scaler.transform(X_pruned)
clf = LogisticRegressionCV(cv=10, scoring='roc_auc', n_jobs=-1)
clf.fit(X_new, y)
print('CV scores: %s' % clf.scores_)
print('Ein: %f' % Util.auc_score(clf, X_new, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('fs', lr),
('scale_new', new_scaler),
('lr', clf)]), 'lr_with_fs1_0703_03')
def erf2():
"""
Submission: erf2_0705_02.csv
3000 trees
E_val: [0.83766072, 0.89704662, 0.85299486, 0.8639041, 0.82955865]
E_in: 1.000000
E_out:
"""
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.ensemble import ExtraTreesClassifier
from sklearn.cross_validation import cross_val_score
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
del X
import gc
gc.collect()
erf = ExtraTreesClassifier(n_estimators=3000, n_jobs=-1,
class_weight='auto')
scores = cross_val_score(erf, X_scaled, y, cv=5, n_jobs=-1)
logger.debug('CV: %s', scores)
logger.debug('Eval: %f', sum(scores) / len(scores))
erf.fit(X_scaled, y)
logger.debug('ExtraTreesClassifier fitted')
logger.debug('Ein: %f', Util.auc_score(erf, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('erf', erf)]), 'erf2_0705_02')
logger.debug('caching fitted ExtraTreesClassifier')
IO.cache(erf, Path.of_cache('erf2.ExtraTreesClassifier.auto.pkl'))
logger.debug('cached fitted ExtraTreesClassifier')
def knn():
"""
Submission: knn_0704_01.csv
E_val:
E_in:
E_out:
"""
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.cross_validation import StratifiedKFold
from sklearn.grid_search import GridSearchCV
from sklearn.neighbors import KNeighborsClassifier
import numpy as np
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
knn = KNeighborsClassifier()
params = {
'n_neighbors': np.arange(5, 51, 5),
'weights': ['uniform', 'distance'],
'leaf_size': np.arange(30, 201, 10)
}
grid = GridSearchCV(knn, params, scoring='roc_auc', n_jobs=-1,
cv=StratifiedKFold(y, 5))
grid.fit(X_scaled, y)
logger.debug('Got best kNN.')
logger.debug('Grid scores: %s', grid.grid_scores_)
logger.debug('Best score (E_val): %s', grid.best_score_)
logger.debug('Best params: %s', grid.best_params_)
IO.cache(grid, Path.of_cache('knn.GridSearchCV.KNeighborsClassifier.pkl'))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('knn', grid)]), 'knn_0704_01')
def lr_with_scale():
"""
Submission: lr_with_scale_0703_01.csv
E_val: <missing>
E_in: 0.878883
E_out: 0.8766589627938616
"""
from sklearn.linear_model import LogisticRegressionCV
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
clf = LogisticRegressionCV(cv=10, scoring='roc_auc', n_jobs=-1)
clf.fit(X_scaled, y)
print('CV scores: %s' % clf.scores_)
print('Ein: %f' % Util.auc_score(clf, X_scaled, y))
IO.dump_submission(Pipeline([('scale_raw', raw_scaler),
('lr', clf)]), 'lr_with_scale_0703_01')
# Convolutional Layer 2. filter_size2 = 3 num_filters2 = 32 # Convolutional Layer 3. filter_size3 = 3 num_filters3 = 64 # Fully-connected layer. fc_size = 128 # Number of neurons in fully-connected layer. # Number of color channels for the images: 1 channel for gray-scale. num_channels = 3 # image dimensions (only squares for now) img_size = 256 # Size of image when flattened to a single dimension img_size_flat = img_size * img_size * num_channels # Tuple with height and width of images used to reshape arrays. img_shape = (img_size, img_size) # class info classes = ['Pool', 'NonPool'] num_classes = len(classes) test_data = dataset.DataSet(*dataset.load_train(input_dir, img_size, classes)) print(test_data.images)
print "Zero-one classification loss", zero_one_loss(y_test_mlb, y_pred)
print "Hamming loss", hamming_loss(y_test_mlb, y_pred)
im = y_test_mlb + y_pred * 2
scipy.misc.imsave("predictions.png", im)
if __name__ == "__main__":
# Load data
print "Loading labels"
label_list = dataset.load_labels()
print "Loading train set"
X_train, y_train, filenames_train = dataset.load_train()
print "Size of train set", len(X_train)
multilabel_classifier(X_train, y_train)
# Unload train set from memory
del X_train, y_train, filenames_train
print "Loading test set"
X_test, y_test, filenames_test = dataset.load_test()
print "Size of test set", len(X_test)
predict(X_test, y_test)
improve_predictions(use_infer_topology=True)
# evaluate_multilabel(y_test, label_list, '../models/pred_ml_improved.pkl')
evaluate_multilabel(y_test, label_list, "../models/pred_ml.pkl")
def gbdt_oob():
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.cross_validation import KFold, train_test_split
import pylab as pl
import numpy as np
X, y = dataset.load_train()
raw_scaler = StandardScaler()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
raw_scaler.fit(X_train)
X_train = raw_scaler.transform(X_train)
X_test = raw_scaler.transform(X_test)
n_estimators = 1000
params = {'n_estimators': n_estimators, 'loss': 'deviance',
'learning_rate': 0.1, 'subsample': 0.5}
gb = GradientBoostingClassifier(**params)
gb.fit(X_train, y_train)
# IO.cache(gb, Path.of_cache('gbdt.GradientBoostingClassifier.pkl'))
logger.debug('Eval: %f', Util.auc_score(gb, X_test, y_test))
logger.debug('Ein: %f', Util.auc_score(gb, X_train, y_train))
x = np.arange(n_estimators) + 1
def heldout_score(clf, X_test, y_test):
"""compute deviance scores on ``X_test`` and ``y_test``. """
score = np.zeros((n_estimators,), dtype=np.float64)
for i, y_pred in enumerate(clf.staged_decision_function(X_test)):
score[i] = clf.loss_(y_test, y_pred)
return score
def cv_estimate(n_folds=3):
cv = KFold(n=X_train.shape[0], n_folds=n_folds)
cv_clf = GradientBoostingClassifier(**params)
val_scores = np.zeros((n_estimators,), dtype=np.float64)
for train, test in cv:
cv_clf.fit(X_train[train], y_train[train])
val_scores += heldout_score(cv_clf, X_train[test], y_train[test])
val_scores /= n_folds
return val_scores
cv_score = cv_estimate(3)
test_score = heldout_score(gb, X_test, y_test)
cumsum = -np.cumsum(gb.oob_improvement_)
oob_best_iter = x[np.argmin(cumsum)]
test_score -= test_score[0]
test_best_iter = x[np.argmin(test_score)]
cv_score -= cv_score[0]
cv_best_iter = x[np.argmin(cv_score)]
oob_color = list(map(lambda x: x / 256.0, (190, 174, 212)))
test_color = list(map(lambda x: x / 256.0, (127, 201, 127)))
cv_color = list(map(lambda x: x / 256.0, (253, 192, 134)))
# IO.cache(cumsum, Path.of_cache('gbdt.cumsum.pkl'))
# IO.cache(test_score, Path.of_cache('gbdt.test_score.pkl'))
# IO.cache(cv_score, Path.of_cache('gbdt.cv_score.pkl'))
pl.plot(x, cumsum, label='OOB loss', color=oob_color)
pl.plot(x, test_score, label='Test loss', color=test_color)
pl.plot(x, cv_score, label='CV loss', color=cv_color)
pl.axvline(x=oob_best_iter, color=oob_color)
pl.axvline(x=test_best_iter, color=test_color)
pl.axvline(x=cv_best_iter, color=cv_color)
xticks = pl.xticks()
xticks_pos = np.array(xticks[0].tolist() +
[oob_best_iter, cv_best_iter, test_best_iter])
xticks_label = np.array(list(map(lambda t: int(t), xticks[0])) +
['OOB', 'CV', 'Test'])
ind = np.argsort(xticks_pos)
xticks_pos = xticks_pos[ind]
xticks_label = xticks_label[ind]
pl.xticks(xticks_pos, xticks_label)
pl.legend(loc='upper right')
pl.ylabel('normalized loss')
pl.xlabel('number of iterations')
pl.savefig('gbdt.oob')
def gbdt2():
"""
Submission: gbdt2_0708_03.csv
n_estimators: 1000, learning_rate: 0.1, subsample: 0.5
E_val: 0.852035
E_in: 0.910251
E_out: 0.8874428893001793
n_estimators: 3000, learning_rate: 0.1, subsample: 0.5
E_val: 0.827988
E_in: 0.938593
E_out: 0.8844206314551558
depth=4; n_estimators: 1000, learning_rate: 0.1, subsample: 0.5
E_val: 0.941602
E_in: 0.983938 (on depth=4) // 0.87209181108731892 (on depth=0)
E_out: 0.8872206627768779
depth=0:
E_val:
E_in: 0.909368 // 0.909368
E_out: 0.8864839071529611
depth=1:
E_val:
E_in: 0.956676 // 0.903537
E_out: 0.8851856544683128
depth=2:
E_val:
E_in: 0.971240 // 0.899843
E_out:
depth=3:
E_val:
E_in: 0.978190 // 0.896956
E_out:
"""
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
import numpy as np
gb = GradientBoostingClassifier(loss='exponential', n_estimators=1000,
learning_rate=0.1, subsample=0.5)
d = 3
X, y = dataset.load_train(depth=d)
raw_scaler = StandardScaler()
raw_scaler.fit(np.r_[X, dataset.load_test()])
X_scaled = raw_scaler.transform(X)
gb.fit(X_scaled, y)
IO.cache(gb, Path.of_cache('gbdt2.GradientBoostingClassifier.d%d.pkl' % d))
IO.dump_submission(Pipeline([('scaler', raw_scaler), ('gbdt', gb)]),
'gbdt2_0708_03.1000.d%d' % d)
logger.debug('E_in(full): %f', Util.auc_score(gb, X_scaled, y))
X, y = dataset.load_train()
X_scaled = raw_scaler.transform(X)
logger.debug('E_in(depth=0): %f', Util.auc_score(gb, X_scaled, y))
print "Zero-one classification loss", zero_one_loss(y_test_mlb, y_pred)
print "Hamming loss", hamming_loss(y_test_mlb, y_pred)
im = y_test_mlb + y_pred * 2
scipy.misc.imsave('predictions.png', im)
if __name__ == '__main__':
#Load data
print "Loading labels"
label_list = dataset.load_labels()
print "Loading train set"
X_train, y_train, filenames_train = dataset.load_train()
print "Size of train set", len(X_train)
multilabel_classifier(X_train, y_train)
#Unload train set from memory
del X_train, y_train, filenames_train
print "Loading test set"
X_test, y_test, filenames_test = dataset.load_test()
print "Size of test set", len(X_test)
predict(X_test, y_test)
improve_predictions(use_infer_topology=True)
#evaluate_multilabel(y_test, label_list, '../models/pred_ml_improved.pkl')
evaluate_multilabel(y_test, label_list, '../models/pred_ml.pkl')