def train(model_name):
#SMOTE for upsampling the minority class
X_train, Y_train = prepare_data(NUM_TRAIN, "data/exoTrain.csv")
sm = SMOTE()
X_train, Y_train = sm.fit_sample(X_train, Y_train)
#Reshape the array from 2D into 3D
X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[1], 1))
model = create_model()
#Add some checkpoints
tensorboard = TensorBoard(log_dir = './Graph', histogram_freq = 0, write_graph = True, write_images = True)
checkpoint_train = ModelCheckpoint(model_path, monitor = "loss", save_best_only = True)
print("Added checkpoints")
model.fit(x = X_train, y = Y_train, epochs = EPOCHS,
callbacks = [tensorboard, checkpoint_train])
# Python 3.7 used # Must have fashion-mnist repo locally import utilities as util import numpy as np from sklearn import metrics from sklearn.linear_model import LogisticRegression def init_logistic_regression(X_train, y_train, X_test, y_test, penalty = 'l2', C = 1.0, max_iter = 5000, solver='lbfgs'): clf = LogisticRegression(penalty=penalty, C=C, max_iter=max_iter, solver=solver) model = clf.fit(X_train, y_train) return model if __name__ == "__main__": # Prep Data x_train, y_train, x_test, y_test, data, target = util.prepare_data() # Plotting the Optimized Model Learning Curve # Set 1: util.learning_curve_plot(LogisticRegression(C=100), data, target, label='C=100', scoring='neg_mean_squared_error', colorTrain='blue', colorTest='magenta') lgr_1 = init_logistic_regression(x_train, y_train, x_test, y_test, C=100, max_iter=5000) pred_1 = lgr_1.predict(x_test) util.print_info(y_test, pred_1) util.learning_curve_plot(LogisticRegression(C=10), data, target, label='C=10', scoring='neg_mean_squared_error', colorTrain='green', colorTest='yellow') lgr_2 = init_logistic_regression(x_train, y_train, x_test, y_test, C=10, max_iter=5000) pred_2 = lgr_2.predict(x_test) util.print_info(y_test, pred_2) util.learning_curve_plot(LogisticRegression(C=1), data, target, label='C=1', scoring='neg_mean_squared_error', colorTrain='cyan', colorTest='red') lgr_3 = init_logistic_regression(x_train, y_train, x_test, y_test, C=1, max_iter=5000) pred_3 = lgr_3.predict(x_test)
RANDOM_STATE = 1
set_random_seed(RANDOM_STATE)
#%% FLC data:
from utilities import prepare_data
from utilities import check_arrays_survival
from flc_data_preprocess import flc_preprocess
#Survival Data
data_x, data_y, protect_attr = flc_preprocess()
# train-test split
data_X_train, data_X_test, data_y_train, data_y_test, S_train, S_test = train_test_split(data_x, data_y, protect_attr, test_size=0.2,stratify=data_y["death"], random_state=7)
data_X_train, data_X_dev, data_y_train, data_y_dev, S_train, S_dev = train_test_split(data_X_train, data_y_train, S_train, test_size=0.2,stratify=data_y_train["death"], random_state=7)
#
data_X_train, data_event_train, data_time_train = check_arrays_survival(data_X_train, data_y_train)
data_X_train, data_event_train, data_time_train, S_train = prepare_data(data_X_train, data_event_train, data_time_train, S_train)
data_X_test, data_event_test, data_time_test = check_arrays_survival(data_X_test, data_y_test)
data_X_test, data_event_test, data_time_test, S_test = prepare_data(data_X_test, data_event_test, data_time_test, S_test)
#
intersectionalGroups = np.unique(S_train,axis=0) # all intersecting groups, i.e. black-women, white-man etc
# data normalization: mean subtraction method to compute euclidean distance
scaler = StandardScaler()
scaler.fit(data_X_train)
data_X_train = scaler.transform(data_X_train)
data_X_test = scaler.transform(data_X_test)
#%%
# hyperparameters of the model
input_size = data_X_train.shape[1]
output_size = 1
def test(predict_or_evaluate, model_path):
#Different results due to loading the model - model is compiled exactly, meaning Dropout still remains
model = load_model(model_path)
X_test, Y_test = prepare_data(NUM_TEST, "data/exoTest.csv")
#Reshape the array from 2D into 3D
X_test = np.reshape(X_test, (X_test.shape[0], X_test.shape[1], 1))
#Predict the inputted images' output
predictions = model.predict(X_test, verbose=1)
print("Predictions: \n" + str(predictions))
if (predict_or_evaluate == "--evaluate"):
#Evaluate the model on test images
evaluations = model.evaluate(X_test, Y_test)
print("Loss: " + str(evaluations[0]))
print("Accuracy: " + str(evaluations[1] * 100) + " %")
true_positives, false_positives, true_negatives, false_negatives = get_positives_and_negatives(
predictions, Y_test)
print("True positive: " + str(true_positives))
print("False positive: " + str(false_positives))
print("True negative: " + str(true_negatives))
print("False negative: " + str(false_negatives))
confusion_matrix = get_confusion_matrix(true_positives,
false_positives,
true_negatives,
false_negatives)
print("Confusion Matrix:\n" + str(confusion_matrix[0]) + "\n" +
str(confusion_matrix[1]))
precision = get_precision(true_positives, false_positives)
print("Precision: " + str(precision))
recall = get_recall(true_positives, false_negatives)
print("Recall/True Positive Rate: " + str(recall))
specificity = get_specificity(false_positives, true_negatives)
print("Specificity/True Negative Rate: " + str(specificity))
F1_score = get_F1(precision, recall)
print("F1 Score: " + str(F1_score))
#False and True Positive Rates specifically for the graph
fpr, tpr, _ = metrics.roc_curve(Y_test, predictions)
roc_auc = metrics.auc(fpr, tpr)
#Graphing the ROC curve
plt.title("ROC Curve for the Exoplanet Detector")
plt.plot(fpr, tpr, "b", label="AUC = %0.2f" % roc_auc)
plt.legend(loc="lower right")
plt.ylabel("True Positive Rate")
plt.xlabel("False Positive Rate")
plt.show()
#Precision and recall for the Precision-Recall curve
precision_graph, recall_graph, _ = metrics.precision_recall_curve(
Y_test, predictions)
auc = metrics.auc(recall_graph, precision_graph)
#Graphing the Precision-Recall Curve
plt.title("Precision-Recall Curve for the Exoplanet Detector")
plt.plot(recall_graph, precision_graph, "b", label="AUC = %0.2f" % auc)
plt.legend(loc="lower right")
plt.xlim([0, 1.1])
plt.ylim([0, 1.1])
plt.ylabel("Precision")
plt.xlabel("Recall")
plt.show()
#Graphing a Confirmed Exoplanet
plt.title("Confirmed Exoplanet")
plt.plot(X_test[0])
plt.ylabel("Light Flux")
plt.xlabel("Time")
plt.show()
#Graphing a Confirmed Non-Exoplanet
plt.title("Confirmed Non-Exoplanet")
plt.plot(X_test[569])
plt.ylabel("Light Flux")
plt.xlabel("Time")
plt.show()