def main():
global nn_input_dim, nn_output_dim, gamma
iters = 15000
hl_nodes = 10
gamma_trials = [0.01, 0.1, 0, 1, 10, 100]
X_train, y_train, X_test, y_test = get_data()
for gamma in gamma_trials:
print "\n-------For gamma = %f and number of hidden layers = %d----------\n" % (
gamma, hl_nodes)
model = train_model(X_train, y_train, hl_nodes, 4, iters, False)
t = predict(X_train, model)
f1 = class_stats(t, y_train, 4)
acc_train.append(f1)
y_pred = predict(X_test, model)
f1 = class_stats(y_pred, y_test, 4)
acc_test.append(f1)
plot_points(gamma_trials, acc_train, 'gamma', 'accuray')
plot_points(gamma_trials, acc_test, 'gamma', 'accuracy')
def perform_multinomial_nb_dir(params):
acc_val = []
for s in [1, 3, 5, 7, 9]:
train_set, train_label, test_set, test_label = create_datasets(s)
y_pred, y_score = predict_multinomial_nb_dir(train_set, train_label,
test_set, params)
plot_pr_curve(test_label, y_score)
acc, _ = class_stats(y_pred, test_label, 2)
acc_val.append(acc)
print sum(acc_val) / 5
y_train = load_data_csv("DS3/train_labels.csv")
X_train_proj = lda_projected_X(X_train, y_train, 1)
# Loading test data and projecting into derived space
X_test = load_data_csv("DS3/test.csv")
y_test = load_data_csv("DS3/test_labels.csv")
X_test_proj = lda_projected_X(X_test, y_test, 1)
# Re-encoding output
y_train[y_train == 1] = 0
y_train[y_train == 2] = 1
y_test[y_test == 1] = 0
y_test[y_test == 2] = 1
# Fitting model
lr = linear_model.LinearRegression()
lr.fit(X_train_proj, y_train)
# Prediction
y_pred_actual = lr.predict(X_test_proj)
y_pred = y_pred_actual.copy()
# Thresholding
y_pred[y_pred < 0.5] = 0
y_pred[y_pred >= 0.5] = 1
# Plotting and printing out results
plot_3d(X_test, y_test)
class_stats(y_pred, y_test, 2)
plot_2d_model(X_test_proj, y_test, lr.coef_[0], lr.intercept_)
def print_performance_metrics(model, X_test, y_test):
print "\n----Model performance metrics------\n"
y_pred = model.predict(X_test)
class_stats(y_pred, y_test, 4)
def perform_multinomial_nb():
train_set, train_label, test_set, test_label = create_datasets(1)
y_pred, y_score = predict_multinomial_nb(train_set, train_label, test_set)
plot_pr_curve(test_label, y_score)
class_stats(y_pred, test_label, 2)