def find_best_params(self,validation_data_x,validation_data_y,n_jobs=1,params=[]):
if not params:
params = self.get_default_param_grid()
merged_x = Data.merge_arrays(self.training_data_x, validation_data_x)
merged_y = Data.merge_arrays(self.training_data_y, validation_data_y)
test_fold = []
for i in range(0,len(self.training_data_y)):
test_fold.append(1)
for i in range(0,len(validation_data_y)):
test_fold.append(0)
cv = PredefinedSplit(test_fold)
gs = GridSearchCV(
estimator=GaussianNB(),
scoring='f1_micro',
param_grid=params,
n_jobs=n_jobs,
cv=cv
)
gs.fit(merged_x,merged_y)
best_params = gs.best_params_
results = gs.cv_results_
return best_params,results
def find_best_params(self,validation_data_x,validation_data_y,alpha_vals,n_jobs=1):
merged_x = Data.merge_arrays(self.training_data_x, validation_data_x)
merged_y = Data.merge_arrays(self.training_data_y, validation_data_y)
test_fold = []
for i in range(0,len(self.training_data_y)):
test_fold.append(1)
for i in range(0,len(validation_data_y)):
test_fold.append(0)
cv = PredefinedSplit(test_fold)
param = {"alpha": alpha_vals}
gs = GridSearchCV(
estimator=BernoulliNB(),
scoring='f1_micro',
param_grid=param,
n_jobs=n_jobs,
cv=cv
)
gs.fit(merged_x,merged_y)
best_params = gs.best_params_
results = gs.cv_results_
return best_params,results
def run_naive_bayes_gaussian():
OUTPUT_PATH = "Outputs/yelp/naive-bayes-fbow-out.txt"
f = open(OUTPUT_PATH, "w")
TRAINING_DATA_PATH = "Data/FrequencyBOW/yelp-train"
VALIDATION_DATA_PATH = "Data/FrequencyBOW/yelp-valid"
TESTING_DATA_PATH = "Data/FrequencyBOW/yelp-test"
f.write(
"Loading Frequency Bag-Of-Words Representation for Training Data\n")
training_data_x = Data.read_x_array(TRAINING_DATA_PATH + "-X.csv")
training_data_y = Data.read_y_array(TRAINING_DATA_PATH + "-Y.csv")
f.write(
"Initializing Gaussian Naive Bayes Class Classifier with training data\n"
)
gnb = GaussianNaiveBayesClassifier(training_data_x=training_data_x,
training_data_y=training_data_y)
f.write(
"Finding best variance smoothing value for Gaussian Naive Bayes Model\n"
)
f.write("Loading validation data\n")
validation_data_x = Data.read_x_array(VALIDATION_DATA_PATH + "-X.csv")
validation_data_y = Data.read_y_array(VALIDATION_DATA_PATH + "-Y.csv")
best_params, results = gnb.find_best_params(validation_data_x,
validation_data_y,
n_jobs=1)
f.write("The best variance smoothing value found was {}\n".format(
best_params["var_smoothing"]))
f.write("\nPerformance metrics for all var_smoothing values tested:\n\n")
index = 0
while (index < 100 and index < len(results['params'])):
f.write("var_smoothing: {} --> {}\n".format(
results['params'][index]['var_smoothing'],
results['mean_test_score'][index]))
index += 1
f.write(
"\n\nInitializing and training a Gaussian Naive Bayes Model with best hyper-parameters\n"
)
gnb = GaussianNaiveBayesClassifier(training_data_x, training_data_y)
gnb.initialize_classifier(var_smoothing=best_params['var_smoothing'])
gnb.train()
testing_data_x = Data.read_x_array(TESTING_DATA_PATH + "-X.csv")
testing_data_y = Data.read_y_array(TESTING_DATA_PATH + "-Y.csv")
f.write("Finding F1-Measure for different datasets\n")
f1_train = gnb.get_f1_measure(training_data_x, training_data_y)
f1_valid = gnb.get_f1_measure(validation_data_x, validation_data_y)
f1_test = gnb.get_f1_measure(testing_data_x, testing_data_y)
f.write("The F1-Measure on training data is {}\n".format(f1_train))
f.write("The F1-Measure on validation data is {}\n".format(f1_valid))
f.write("The F1-Measure on testing data is {}\n".format(f1_test))
f.close()
result.columns = np.concatenate(
[ask_column, bid_column, ask_vol_column, bid_vol_column, mid_column])
result.iloc[:, :20] = result.iloc[:, :20].values / result[
'mid'].values[:, np.newaxis] - 1
for i in range(9):
result.iloc[:, i + 21] += result.iloc[:, i + 20]
result.iloc[:, i + 31] += result.iloc[:, i + 30]
result = result.iloc[1:]
labels = [1, 2, 3, 5, 10]
label = data.iloc[:, labels.index(k) - 5]
return result, label
if __name__ == '__main__':
from data_processor import DataForModel as Data
data = DataGenerator(test=True)
data, label = Preprocess(data)
data = pd.concat([data, label], axis=1)
print(data.shape)
data = Data(data, test_ratio=1)
print(data.len_train)
x, y = data.get_test_batch(300, False)
print(x.shape)
def run_naive_bayes_bernoulli():
OUTPUT_PATH = "Outputs/yelp/naive-bayes-bbow-out.txt"
f = open(OUTPUT_PATH, "w+")
TRAINING_DATA_PATH = "Data/BinaryBOW/yelp-train"
VALIDATION_DATA_PATH = "Data/BinaryBOW/yelp-valid"
TESTING_DATA_PATH = "Data/BinaryBOW/yelp-test"
f.write("Loading Binary Bag-Of-Words Representation for Training Data\n")
training_data_x = Data.read_x_array(TRAINING_DATA_PATH + "-X.csv")
training_data_y = Data.read_y_array(TRAINING_DATA_PATH + "-Y.csv")
f.write(
"Initializing Bernoulli Naive Bayes Class Classifier with training data\n"
)
bnb = BernoulliNaiveBayesClassifier(training_data_x=training_data_x,
training_data_y=training_data_y)
f.write("Finding best alpha value for Naive Bayes Bernoulli Model\n")
f.write("Loading validation data\n")
validation_data_x = Data.read_x_array(VALIDATION_DATA_PATH + "-X.csv")
validation_data_y = Data.read_y_array(VALIDATION_DATA_PATH + "-Y.csv")
alpha_start = 0
alpha_stop = 1
num_intervals = 200
alpha_vals = np.linspace(start=alpha_start,
stop=alpha_stop,
num=num_intervals)
f.write("Testing {} alpha values between {} and {}:\n".format(
num_intervals, alpha_start, alpha_stop))
best_params, results = bnb.find_best_params(validation_data_x,
validation_data_y,
alpha_vals,
n_jobs=10)
f.write("The best alpha value found was {}\n".format(best_params["alpha"]))
f.write("\nPerformance metrics for all alpha values tested:\n\n")
output = ""
for i in range(0, len(alpha_vals)):
output += "Alpha value: " + str(
alpha_vals[i]) + " --> F1-Score: " + str(
results['mean_test_score'][i]) + "\n"
f.write(output)
f.write(
"\n\nInitializing and training a Bernoulli Naive Bayes Model with alpha={}\n"
.format(best_params['alpha']))
alpha = float(best_params['alpha'])
bnb = BernoulliNaiveBayesClassifier(training_data_x, training_data_y,
alpha)
bnb.train()
testing_data_x = Data.read_x_array(TESTING_DATA_PATH + "-X.csv")
testing_data_y = Data.read_y_array(TESTING_DATA_PATH + "-Y.csv")
f.write("Finding F1-Measure for different datasets\n")
f1_train = bnb.get_f1_measure(training_data_x, training_data_y)
f1_valid = bnb.get_f1_measure(validation_data_x, validation_data_y)
f1_test = bnb.get_f1_measure(testing_data_x, testing_data_y)
f.write("The F1-Measure on training data with alpha={} is {}\n".format(
alpha, f1_train))
f.write("The F1-Measure on validation data with alpha={} is {}\n".format(
alpha, f1_valid))
f.write("The F1-Measure on testing data with alpha={} is {}\n".format(
alpha, f1_test))
f.close()
def run_decision_tree():
OUTPUT_PATH = "Outputs/yelp/decision-tree-bbow-out.txt"
f = open(OUTPUT_PATH, "w")
TRAINING_DATA_PATH = "Data/BinaryBOW/yelp-train"
VALIDATION_DATA_PATH = "Data/BinaryBOW/yelp-valid"
TESTING_DATA_PATH = "Data/BinaryBOW/yelp-test"
f.write("Loading Binary Bag-Of-Words Representation for Training Data\n")
training_data_x = Data.read_x_array(TRAINING_DATA_PATH + "-X.csv")
training_data_y = Data.read_y_array(TRAINING_DATA_PATH + "-Y.csv")
f.write("Initializing Decision Tree Classifier with training data\n")
dt = DecisionTree(training_data_x=training_data_x,
training_data_y=training_data_y)
f.write("Loading validation data\n")
validation_data_x = Data.read_x_array(VALIDATION_DATA_PATH + "-X.csv")
validation_data_y = Data.read_y_array(VALIDATION_DATA_PATH + "-Y.csv")
f.write("Finding the best hyper-parameters:\n")
best_params, best_score, results = dt.find_best_params(validation_data_x,
validation_data_y,
n_jobs=1)
f.write("The best hyper-parameters are as follows: \n")
f.write(
"max_depth: {}\t| min_samples_split: {}\t| min_samples_leaf: {}\t| max_features: {} with an average F1-Measure of {}\n\n"
.format(best_params['max_depth'], best_params['min_samples_split'],
best_params['min_samples_leaf'], best_params['max_features'],
best_score))
f.write(
"\nPerformance metrics for the first 100 hyper-parameters_tested:\n\n")
index = 0
while (index < 100 and index < len(results['params'])):
f.write(
"max_depth: {}\t| min_samples_split: {}\t| min_samples_leaf: {}\t| max_features: {} --> {}\n"
.format(results['params'][index]['max_depth'],
results['params'][index]['min_samples_split'],
results['params'][index]['min_samples_leaf'],
results['params'][index]['max_features'],
results['mean_test_score'][index]))
index += 1
f.write(
"\n\nInitializing and training a Decision Tree Classifier with the best parameters \n"
)
dt = DecisionTree(training_data_x, training_data_y)
dt.initialize_classifier(
max_depth=best_params['max_depth'],
min_samples_split=best_params['min_samples_split'],
min_samples_leaf=best_params['min_samples_leaf'],
max_features=best_params['max_features'])
dt.train()
testing_data_x = Data.read_x_array(TESTING_DATA_PATH + "-X.csv")
testing_data_y = Data.read_y_array(TESTING_DATA_PATH + "-Y.csv")
f.write("Finding F1-Measure for different datasets\n")
f1_train = dt.get_f1_measure(training_data_x, training_data_y)
f1_valid = dt.get_f1_measure(validation_data_x, validation_data_y)
f1_test = dt.get_f1_measure(testing_data_x, testing_data_y)
f.write(
"The F1-Measure on training data with these parameters is {}\n".format(
f1_train))
f.write("The F1-Measure on validation data with these parameters is {}\n".
format(f1_valid))
f.write(
"The F1-Measure on testing data with these parameters is {}\n".format(
f1_test))
f.close()
def run_linear_svm():
OUTPUT_PATH = "Outputs/yelp/linear-svm-bbow-out.txt"
f = open(OUTPUT_PATH, "w")
TRAINING_DATA_PATH = "Data/BinaryBOW/yelp-train"
VALIDATION_DATA_PATH = "Data/BinaryBOW/yelp-valid"
TESTING_DATA_PATH = "Data/BinaryBOW/yelp-test"
f.write("Loading Binary Bag-Of-Words Representation for Training Data\n")
training_data_x = Data.read_x_array(TRAINING_DATA_PATH + "-X.csv")
training_data_y = Data.read_y_array(TRAINING_DATA_PATH + "-Y.csv")
f.write(
"Initializing Linear Support Vector Classifier with training data\n")
lsvc = LinearSupportVectorClassifier(training_data_x=training_data_x,
training_data_y=training_data_y)
f.write("Loading validation data\n")
validation_data_x = Data.read_x_array(VALIDATION_DATA_PATH + "-X.csv")
validation_data_y = Data.read_y_array(VALIDATION_DATA_PATH + "-Y.csv")
f.write("Finding the best hyper-parameters:\n")
best_params, best_score, results = lsvc.find_best_params(validation_data_x,
validation_data_y,
n_jobs=10)
f.write("The best hyper-parameters are as follows: \n")
f.write("C: {}\t| tol: {} with an F1-Measure of {}\n\n".format(
best_params['C'], best_params['tol'], best_score))
f.write(
"\nPerformance metrics for the first 100 hyper-parameters_tested:\n\n")
index = 0
while (index < 100 and index < len(results['params'])):
f.write("C: {}\t| tol: {} --> {}\n".format(
results['params'][index]['C'], results['params'][index]['tol'],
results['mean_test_score'][index]))
index += 1
f.write(
"\n\nInitializing and training a Linear Support Vector Classifier with C={} and tol={} \n"
.format(best_params['C'], best_params['tol']))
best_C = float(best_params['C'])
best_tol = float(best_params['tol'])
lsvc = LinearSupportVectorClassifier(training_data_x, training_data_y)
lsvc.initialize_classifier(tol=best_tol, C=best_C)
lsvc.train()
testing_data_x = Data.read_x_array(TESTING_DATA_PATH + "-X.csv")
testing_data_y = Data.read_y_array(TESTING_DATA_PATH + "-Y.csv")
f.write("Finding F1-Measure for different datasets\n")
f1_train = lsvc.get_f1_measure(training_data_x, training_data_y)
f1_valid = lsvc.get_f1_measure(validation_data_x, validation_data_y)
f1_test = lsvc.get_f1_measure(testing_data_x, testing_data_y)
f.write(
"The F1-Measure on training data with C={} and tol={} is {}\n".format(
best_C, best_tol, f1_train))
f.write("The F1-Measure on validation data with C={} and tol={} is {}\n".
format(best_C, best_tol, f1_valid))
f.write(
"The F1-Measure on testing data with C={} and tol={} is {}\n".format(
best_C, best_tol, f1_test))
f.close()