def find_best_parameters_one_hot(features, labels, encoded_labels,
num_classes):
results = []
for distance_function in parameters.distance_functions:
for kernel_function in parameters.kernel_functions:
for window_type in parameters.window_types:
for window_parameter in parameters.window_widths if window_type == 'fixed' else parameters.window_neighbours:
predicted_labels = []
for index in range(len(features)):
predicted_label_values = []
for label_index in range(len(encoded_labels[index])):
predicted_label_value = classify_object(
features[np.arange(len(features)) != index],
encoded_labels[np.arange(len(encoded_labels))
!= index][:, label_index],
features[index], parameters.
best_combination['distance_function'],
parameters.best_combination['kernel_function'],
parameters.best_combination['window_type'],
window_parameter)
predicted_label_values.append(
predicted_label_value)
predicted_value = [
i for i, j in enumerate(predicted_label_values)
if j == max(predicted_label_values)
][0]
predicted_label = round(predicted_value)
predicted_labels.append(predicted_label)
f_score = f1_score(labels,
predicted_labels,
labels=[i for i in range(num_classes)],
average='weighted')
results.append({
'f_score': f_score,
'distance_function': distance_function,
'kernel_function': kernel_function,
'window_type': window_type,
'window_parameter': window_parameter
})
sorted_results = sorted(results,
key=lambda configuration: configuration['f_score'],
reverse=True)
return sorted_results
def find_best_parameters_native(features, labels, num_classes):
results = []
for distance_function in distance_functions:
for kernel_function in kernel_functions:
for window_type in window_types:
for window_parameter in parameters.window_widths if window_type == 'fixed' else parameters.window_neighbours:
# count every existent combination of parameters and get F-score for it
predicted_labels = []
for index in range(len(features)):
# get predicted value of features[index] (of every feature actually) and set label (label if class value) as round of predicted
predicted_value = classify_object(
features[np.arange(len(features)) != index],
labels[np.arange(len(labels)) != index],
features[index], distance_function,
kernel_function, window_type, window_parameter)
predicted_label = round(predicted_value)
predicted_labels.append(predicted_label)
f_score = f1_score(labels,
predicted_labels,
labels=[i for i in range(num_classes)],
average='weighted')
results.append({
'f_score': f_score,
'distance_function': distance_function,
'kernel_function': kernel_function,
'window_type': window_type,
'window_parameter': window_parameter
})
# after i have got all the results, decide which f-score is better by its value (the bigger the better)
sorted_results = sorted(results,
key=lambda configuration: configuration['f_score'],
reverse=True)
return sorted_results
data_csv = './data/glass.csv'
dataset, features, labels, num_classes = prepare_data(data_csv)
#print(find_best_parameters(features, labels, num_classes))
f_scores = []
for window_parameter in parameters.window_neighbours:
predicted_labels = []
for index in range(len(features)):
predicted_value = classify_object(
features[np.arange(len(features)) != index],
labels[np.arange(len(labels)) != index], features[index],
parameters.best_combination['distance_function'],
parameters.best_combination['kernel_function'],
parameters.best_combination['window_type'], window_parameter)
predicted_label = round(predicted_value)
predicted_labels.append(predicted_label)
f_score = f1_score(labels,
predicted_labels,
labels=[i for i in range(num_classes)],
average='weighted')
f_scores.append({'f_score': f_score, 'window_parameter': window_parameter})
print(f_scores)
plt.plot([point['window_parameter'] for point in f_scores],