def optimization(x, y, x_test, y_test):
print("Multilayer perceptron optimizer")
best_result = -999999
a = 0 # 80 or 70
b = 0 # 40 2.5496 % 25 also 3.1%
for i in range(50, 80, 10):
for j in range(5, 45, 10):
print("i: {0}, j: {1}".format(i, j))
mlp = MLPClassifier(hidden_layer_sizes=(i, j))
mlp.fit(x, y)
y_pred = mlp.predict(x_test)
confusion = confusion_matrix(y_test, y_pred)
result = judge.get_points(confusion)
if result > best_result:
best_result = result
a = i
b = j
print("Best so far: a:{0}, b:{1}".format(a, b))
new_a = a
new_b = b
range_to_search_a = 5
range_to_search_b = 3
for i in range(new_a - range_to_search_a, new_a + range_to_search_a, 1):
for j in range(new_b - range_to_search_b, new_b + range_to_search_b,
1):
print("i: {0}, j: {1}".format(i, j))
mlp = MLPClassifier(hidden_layer_sizes=(i, j))
mlp.fit(x, y)
y_pred = mlp.predict(x_test)
confusion = confusion_matrix(y_test, y_pred)
result = judge.get_points(confusion)
if result > best_result:
best_result = result
a = i
b = j
print("Found best: a:{0}, b:{1}".format(a, b))
return [a, b]
def optimize(x, y, x_test, y_test):
best_result = -99999999
best_number = 1
for i in range(1, 20, 1):
print(i)
knn = KNeighborsClassifier(i)
knn.fit(x, y)
y_pred = knn.predict(x_test)
confusion = confusion_matrix(y_test, y_pred)
result = judge.get_points(confusion)
if result > best_result:
best_result = result
best_number = i
print("Best neighbors number = {0}".format(best_number))
return best_number
def display(confusion, labels, number, name=""):
print("Displaying!")
cm = confusion
print(cm)
plt.figure(number)
ax = plt.subplot()
sns.heatmap(cm, annot=True, ax=ax, fmt='g')
#annot=True to annotate cells
# labels, title and ticks
ax.set_xlabel('Predicted labels')
ax.set_ylabel('True labels')
result = judge.get_points(confusion)
result = -result * 100
result = round(result, 4)
if result > 100:
result = "Just really bad.."
ax.set_title(
"Confusion matrix of the {0} classifier, \nresult: {1} % of bad classification"
.format(name, result))
ax.xaxis.set_ticklabels(labels)
ax.yaxis.set_ticklabels(labels)
def error_scorer(y_true, y_pred):
confusion = confusion_matrix(y_true, y_pred)
return judge.get_points(confusion)