results = {}
# SVM ensemble
print("\nSVM ENSEMBLE:")
from sklearn.svm import SVC
model = SVC(verbose=True, max_iter=1e5)
params = {"kernel": ["linear", "rbf", "poly", "sigmoid"], "C": [0.1, 1, 10]}
ensemble_model = fit_ensemble(members, ensembleX_train, y_train, train_size,
model, params)
# grid search results
ensemble_results = pd.DataFrame.from_dict(ensemble_model.cv_results_)
print(ensemble_results)
ensemble_results.to_csv("../../results/ensembles_svm.csv")
print("Best model score:", ensemble_model.best_score_)
print(ensemble_model.best_params_)
# predict output using ensemble model
yhat_train = ensemble_predict(members, ensemble_model, ensembleX_train,
train_size)
train_acc = accuracy_score(y_train, yhat_train)
print("Train accuracy:", train_acc)
yhat_test = ensemble_predict(members, ensemble_model, ensembleX_test,
test_size)
test_acc = accuracy_score(y_test, yhat_test)
print("Test accuracy:", test_acc)
results["SVM(E)"] = {"Train": train_acc, "Test": test_acc}
# results to data frame
df = pd.DataFrame.from_dict(results, orient="index")
df = df.reset_index()
df = df.rename(columns={
"index": "Model",
"Train": "Train Accuracy",
"Test": "Test Accuracy"
model_names = ['Xception', 'DenseNet121', 'MobileNetV2']
models = []
for name in model_names:
print("Model:", name)
m = get_model(name, **config['model_params'])
m.load_weights('./models/from-scratch/' + name + '.h5')
# m.compile(loss='binary_crossentropy', optimizer=tf.keras.optimizers.Adam(), metrics=['accuracy', tf.keras.metrics.AUC()])
# m.evaluate(test_dataset)
models.append(m)
# get numpy arrays from tensorflow datasets
y_test = tf_dataset_to_numpy(test_dataset)
y_val = tf_dataset_to_numpy(val_dataset)
# perform ensemble prediction
y_probs = ensemble_predict(models, val_dataset)
y_pred = (y_probs > 0.5).astype(np.uint8).flatten()
# print the ensemble eval metrics
get_eval_scores(y_val, y_probs, y_pred)
class_names = np.array(['normal', 'tumor'])
# Plot non-normalized confusion matrix
plot_confusion_matrix(y_val,
y_pred,
classes=class_names,
title='Confusion matrix')
plt.show()
########################
print("----------Test----------")