Beispiel #1
0
def log(info):
    f = file('submission.txt', 'w+')
    f.write(str(info))
    f.close()

def submission(preds):
    out = ""
    for p in preds:
        out += str(p) + "\n"
    log(out)


data, targets = Data.data()
print "training data: ", len(data)
test = Data.test()
print "test data: ", len(test)
data = data + test
print "all data: ", len(data)

# preprocessing
start = time()
matrix = BlackboxPreprocess.to_matrix(data)
print matrix.shape
matrix = BlackboxPreprocess.scale(matrix)
#matrix = BlackboxPreprocess.polynomial(matrix, 2)
matrix = preprocessing.normalize(matrix, norm='l2')
min_max_scaler = preprocessing.MinMaxScaler(feature_range=(-1.,1.))
matrix = min_max_scaler.fit_transform(matrix)
#matrix = BlackboxPreprocess.norm(matrix)
print matrix.shape
Beispiel #2
0
from preprocess import Polynomial
from abstract_model import ClassifierEnsemble
from sklearn_wrapper import LinearRegressionModel, LogisticRegressionModel, SVCModel
from score import score
import logging


"""
This is supposed to be a sample interface for me to develop to support.
"""

objective = Objective.MAXIMIZE

# data
train_data, train_targets = Data.train()
test_data, test_targets = Data.test()

# feature engineering
extra_data = test_data
pipe = Pipeline(Polynomial, LogisticRegressionModel, objective, logging.WARN)
pipe.fit(train_data, train_targets, extra_data)
print pipe.hyperparams

# train model
train_data = pipe.transform(train_data)

voter1 = LogisticRegressionModel(objective, logging.INFO)
models = [m(objective, logging.INFO) for m in [SVCModel, LogisticRegressionModel]]
ensemble = ClassifierEnsemble(models, voter1, objective, logging.INFO)

voter2 = LogisticRegressionModel(objective, logging.INFO)
from data import Data
from blackbox_preprocess import BlackboxPreprocess
from sklearn.linear_model import LogisticRegression


data, targets = Data.data()
extra = Data.test()
data = data + extra
originals = data

# preprocessing
matrix = BlackboxPreprocess.to_matrix(data)
print "(examples, dimensions): ", matrix.shape
matrix = BlackboxPreprocess.scale(matrix)
matrix = BlackboxPreprocess.polynomial(matrix, 2)
print "(examples, dimensions): ", matrix.shape
data = matrix.tolist()

# split training and CV data
tr_data = data[:1000]
unlabeled = data[1000:]

# create psuedo labels
model = LogisticRegression(C=1.3, penalty='l1', tol=0.05)
print len(targets)
print targets[:10]
model.fit(tr_data, targets)

labeled = []
for i,u in enumerate(unlabeled):
    orig = originals[i]