def main():
# Load processed data
# You could use the following script to generate a well-processed train and test data sets:
# https://www.kaggle.com/yassinealouini/predicting-red-hat-business-value/features-processing
# I have only used the .head() of the data sets since the process takes a long time to run.
# I have also put the act_train and act_test data sets since I don't have the processed data sets
# loaded.
print("Loading data...")
tour = 89
nrows = 1000
nrows = None
# Load the data from the CSV files
training_data = data_helper.load_training_data(tour, nrows=nrows)
# training_data = training_data.sample(frac=0.1)
#prediction_data = data_helper.load_testing_data(tour, nrows=nrows)
X, y = data_helper.get_Xy(training_data)
#-------------------------------------------------#
# Extract the train and valid (used for validation) dataframes from the train_df
#-------------------------------------------------#
# Run the optimization
# Trials object where the history of search will be stored
# For the time being, there is a bug with the following version of hyperopt.
# You can read the error messag on the log file.
# For the curious, you can read more about it here: https://github.com/hyperopt/hyperopt/issues/234
# => So I am commenting it.
trials = Trials()
# score = make_scorer(X, y)
score = make_cv_scorer(X, y, cv=5)
best_hyperparams = optimize(score, trials, max_evals=50)
print("The best hyperparameters are:")
print(best_hyperparams)
print('trials:')
for t in trials.trials:
print(t)
def train(nn):
nrows = 30
nrows = None
training_data = data_helper.load_training_data(tour, nrows=nrows)
# training_data = data_helper.get_random_data(nrows=5, nfeat=3)
# training_data = data_helper.get_random_data()
frac = None
#frac = 0.1
#frac = 0.01
#frac = 0.0001
if frac is not None:
training_data = training_data.sample(frac=frac)
X, y = data_helper.get_Xy(training_data, onehot=True)
batch_count = 50000
batch_size = 5000
train_model(nn, X, y, batch_size=batch_size, batch_count=batch_count)
def main():
# Set seed for reproducibility
np.random.seed(0)
print("Loading data...")
tour = 90
nrows = None
# Load the data from the CSV files
training_data = data_helper.load_training_data(tour, nrows=nrows)
#training_data = pd.read_csv(data_dir + '/numerai_training_data.csv', header=0)
#prediction_data = pd.read_csv(data_dir + '/numerai_tournament_data.csv', header=0)
#prediction_data = data_helper.load_testing_data(tour, nrows=nrows)
features = data_helper.get_feature_names(training_data)
# Transform the loaded CSV data into numpy arrays
frac = 0.1
frac = 0.01
frac = 0.0001
frac = None
if frac is not None:
training_data = training_data.sample(frac=frac)
X, y = data_helper.get_Xy(training_data)
xgb1 = xgb.XGBClassifier(
learning_rate =0.1,
n_estimators=1000,
max_depth=5,
min_child_weight=1,
gamma=0,
subsample=0.8,
colsample_bytree=0.8,
objective='binary:logistic',
nthread=4,
scale_pos_weight=1,
seed=27)
#modelfit(xgb1, X, y)
tune1(xgb1, X, y)
import tensorflow as tf
#import numpy as np
import seq2seq
import pickle
#from tensorflow.python.layers.core import Dense
#%%
## first, load and pad data
## load all data and vocabulary
vocab_path = os.path.join(config.PROCESSED_PATH, 'vocab.p')
train_token_path = os.path.join(config.PROCESSED_PATH, 'processed_tokens.p')
vocab_to_int, int_to_vocab = helper.load_vocab(vocab_path)
config.source_vocab_size = len(vocab_to_int)
config.target_vocab_size = len(vocab_to_int)
train_enc_tokens, train_dec_tokens, test_enc_tokens, test_dec_tokens = helper.load_training_data(
train_token_path)
bucket_ids = helper.bucket_training_data(train_enc_tokens,
config.max_conv_length)
batches = helper.make_batches_of_bucket_ids(bucket_ids, config.batch_size)
## get a batch of data nd pad them
#%%
## build the network
# create inpute place holder
input_data, targets, lr, keep_prob, target_sequence_length, max_target_sequence_length, source_sequence_length, hrnn_sequence_length = seq2seq.model_inputs(
)
# get input shape
input_shape = tf.shape(input_data)
def main():
# Set seed for reproducibility
np.random.seed(0)
print("Loading data...")
nrows = None
# Load the data from the CSV files
training_data = data_helper.load_training_data(tour, nrows=nrows)
#training_data = pd.read_csv(data_dir + '/numerai_training_data.csv', header=0)
#prediction_data = pd.read_csv(data_dir + '/numerai_tournament_data.csv', header=0)
prediction_data = data_helper.load_testing_data(tour, nrows=nrows)
features = data_helper.get_feature_names(training_data)
# Transform the loaded CSV data into numpy arrays
frac = 0.1
frac = 0.01
frac = 0.0001
frac = None
if frac is not None:
training_data = training_data.sample(frac=frac)
X, y = data_helper.get_Xy(training_data)
# This is your model that will learn to predict
rfc = RandomForestClassifier(
#max_depth=None,
max_depth=10,
n_estimators=20,
max_features=1,
min_samples_split=500,
min_samples_leaf=50,
n_jobs=1)
adbr = AdaBoostClassifier()
#clf = svm.SVC(kernel='poly', gamma=1, probability=True)
svc_linear = clf = svm.SVC(kernel='linear', gamma=0.1, probability=True)
#clf = svm.SVC(kernel='rbf', gamma=1, probability=True)
#clf = Pipeline([('poly', PolynomialFeatures(degree=2, include_bias=False)),
# ('logistic', linear_model.LogisticRegression(fit_intercept=True))])
lr = linear_model.LogisticRegression()
lr1 = linear_model.LogisticRegression(C=1000)
xgbr = XGBClassifier(n_jobs=2)
xgbr_linear = XGBClassifier(n_jobs=2, booster='gblinear')
xgbr2 = XGBClassifier(max_depth=3, n_estimators=200)
xgbr_opt_param = {
'eval_metric': 'logloss',
'objective': 'binary:logistic',
# Increase this number if you have more cores. Otherwise, remove it and it will default
# to the maxium number.
'nthread': 2,
'booster': 'gbtree',
'tree_method': 'exact',
'silent': 1,
'seed': SEED,
}
tmp = {
'eta': 0.07500000000000001,
'n_estimators': 21.0,
'gamma': 0.75,
'colsample_bytree': 0.8500000000000001,
'min_child_weight': 4.0,
'subsample': 0.75,
'max_depth': 3
}
xgbr_opt_param.update(tmp)
xgbr_opt = XGBClassifier(**(xgboost_param_convert(xgbr_opt_param)))
#model = (Cutter(method='qcut', nbins=100),
# OneHotEncoder(handle_unknown='ignore'))
#model += (clf,)
models = [
#('logistic', lr),
#('logistic-C1000', lr1),
#('linear SVC', svc_linear), too slow
#('randomforest', rfc),
#('adaboost', adbr),
#('xgbr-default', xgbr),
('xgbr-linear', xgbr_linear),
# ('xgbr-200', xgbr2),
# ('xgbr-opt', xgbr_opt),
]
perfs = train_eval_model(models, X, y, prediction_data, features=features)
print_perfs(models, perfs)
