def mlp(x, y, x_test, hidden_units=128):
mlp = MLPClassifier(hidden_layer_sizes=(hidden_units, hidden_units,
hidden_units),
max_iter=1000,
solver='lbfgs')
scaler = preprocessing.StandardScaler()
x = scaler.fit_transform(x)
x_test = scaler.transform(x_test)
mlp.fit(x, y)
return mlp._predict(x_test)[:,
0], get_best_threshold(y,
mlp._predict(x)[:, 0])
# load and train
iris = datasets.load_iris()
X = iris.data
y = iris.target
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
num_inputs = X.shape[1]
hidden_layers = (2, 2) #10, 10, 10 would be 3 layers of 10
model = MLPClassifier(activation='identity',
hidden_layer_sizes=hidden_layers,
max_iter=4000,
random_state=1)
model.out_activation_ = "identity"
model.fit(X_train, y_train)
predictions = model.predict(X_test)
pred = model._predict(X_test)
error = (y_test != predictions).sum()
print("Trained Neural Network with accuracy of " +
str(100 - (error / len(predictions))) + ".\n")
weights = np.array(model.coefs_)
bias = np.array(model.intercepts_)
temp_left_nodes = []
for m in range(len(X_test)):
left_nodes = [(x) for x in X_test[m]]
#For each layer of weights
for i in range(weights.shape[0]):
for k in range(weights[i].shape[1]):
sum = None
# for each ROW of the matrix