def fit(self, x, Y):
self.classes = np.sort(list(set(Y)))
x = np.array(x).copy()
y = np.array(Y).copy()
for f in range(len(self.classes)):
y[y == self.classes[f]] = f
clf = MLPClassifier(hidden_layer_sizes=self.shape,
alpha=self.beta,
learning_rate_init=self.eta0)
clf.partial_fit(x, Y, classes=self.classes)
coefs = clf.coefs_
intercepts = clf.intercepts_
coefs_tot = [np.zeros(np.shape(c)) for c in coefs]
intercepts_tot = [np.zeros(np.shape(c)) for c in intercepts]
for f in range(self.agg):
if self.verbose and (self.agg > 1):
print('Passage ' + str(f))
clfi = self.fit1(x, Y, clf, coefs, intercepts)
for i in range(len(coefs)):
coefs_tot[i] += clfi.coefs_[i]
intercepts_tot[i] += clfi.intercepts_[i]
coefs = [c / self.agg for c in coefs_tot]
intercepts = [c / self.agg for c in intercepts_tot]
clf.coefs_ = coefs
clf.intercepts_ = intercepts
self.clf = clf
def generate_NN():
data_set = load_breast_cancer()
X = data_set['data']
y = data_set['target']
X_train, X_test, y_train, y_test = train_test_split(X, y)
scaler = StandardScaler()
scaler.fit(X_train)
X_train = scaler.transform(X_train)
X_test = scaler.transform(X_test)
mlp = MLPClassifier(hidden_layer_sizes=hidden_layers_size)
mlp.fit(X_train, y_train)
mlp.intercepts_ = get_zeroes_biases_vectors(input_layer_size,
hidden_layers_size,
output_layer_size)
return X_test, mlp, y_test
def test_fit():
# Test that the algorithm solution is equal to a worked out example.
X = np.array([[0.6, 0.8, 0.7]])
y = np.array([0])
mlp = MLPClassifier(solver='sgd', learning_rate_init=0.1, alpha=0.1,
activation='logistic', random_state=1, max_iter=1,
hidden_layer_sizes=2, momentum=0)
# set weights
mlp.coefs_ = [0] * 2
mlp.intercepts_ = [0] * 2
mlp.n_outputs_ = 1
mlp.coefs_[0] = np.array([[0.1, 0.2], [0.3, 0.1], [0.5, 0]])
mlp.coefs_[1] = np.array([[0.1], [0.2]])
mlp.intercepts_[0] = np.array([0.1, 0.1])
mlp.intercepts_[1] = np.array([1.0])
mlp._coef_grads = [] * 2
mlp._intercept_grads = [] * 2
# Initialize parameters
mlp.n_iter_ = 0
mlp.learning_rate_ = 0.1
# Compute the number of layers
mlp.n_layers_ = 3
# Pre-allocate gradient matrices
mlp._coef_grads = [0] * (mlp.n_layers_ - 1)
mlp._intercept_grads = [0] * (mlp.n_layers_ - 1)
mlp.out_activation_ = 'logistic'
mlp.t_ = 0
mlp.best_loss_ = np.inf
mlp.loss_curve_ = []
mlp._no_improvement_count = 0
mlp._intercept_velocity = [np.zeros_like(intercepts) for
intercepts in
mlp.intercepts_]
mlp._coef_velocity = [np.zeros_like(coefs) for coefs in
mlp.coefs_]
mlp.partial_fit(X, y, classes=[0, 1])
# Manually worked out example
# h1 = g(X1 * W_i1 + b11) = g(0.6 * 0.1 + 0.8 * 0.3 + 0.7 * 0.5 + 0.1)
# = 0.679178699175393
# h2 = g(X2 * W_i2 + b12) = g(0.6 * 0.2 + 0.8 * 0.1 + 0.7 * 0 + 0.1)
# = 0.574442516811659
# o1 = g(h * W2 + b21) = g(0.679 * 0.1 + 0.574 * 0.2 + 1)
# = 0.7654329236196236
# d21 = -(0 - 0.765) = 0.765
# d11 = (1 - 0.679) * 0.679 * 0.765 * 0.1 = 0.01667
# d12 = (1 - 0.574) * 0.574 * 0.765 * 0.2 = 0.0374
# W1grad11 = X1 * d11 + alpha * W11 = 0.6 * 0.01667 + 0.1 * 0.1 = 0.0200
# W1grad11 = X1 * d12 + alpha * W12 = 0.6 * 0.0374 + 0.1 * 0.2 = 0.04244
# W1grad21 = X2 * d11 + alpha * W13 = 0.8 * 0.01667 + 0.1 * 0.3 = 0.043336
# W1grad22 = X2 * d12 + alpha * W14 = 0.8 * 0.0374 + 0.1 * 0.1 = 0.03992
# W1grad31 = X3 * d11 + alpha * W15 = 0.6 * 0.01667 + 0.1 * 0.5 = 0.060002
# W1grad32 = X3 * d12 + alpha * W16 = 0.6 * 0.0374 + 0.1 * 0 = 0.02244
# W2grad1 = h1 * d21 + alpha * W21 = 0.679 * 0.765 + 0.1 * 0.1 = 0.5294
# W2grad2 = h2 * d21 + alpha * W22 = 0.574 * 0.765 + 0.1 * 0.2 = 0.45911
# b1grad1 = d11 = 0.01667
# b1grad2 = d12 = 0.0374
# b2grad = d21 = 0.765
# W1 = W1 - eta * [W1grad11, .., W1grad32] = [[0.1, 0.2], [0.3, 0.1],
# [0.5, 0]] - 0.1 * [[0.0200, 0.04244], [0.043336, 0.03992],
# [0.060002, 0.02244]] = [[0.098, 0.195756], [0.2956664,
# 0.096008], [0.4939998, -0.002244]]
# W2 = W2 - eta * [W2grad1, W2grad2] = [[0.1], [0.2]] - 0.1 *
# [[0.5294], [0.45911]] = [[0.04706], [0.154089]]
# b1 = b1 - eta * [b1grad1, b1grad2] = 0.1 - 0.1 * [0.01667, 0.0374]
# = [0.098333, 0.09626]
# b2 = b2 - eta * b2grad = 1.0 - 0.1 * 0.765 = 0.9235
assert_almost_equal(mlp.coefs_[0], np.array([[0.098, 0.195756],
[0.2956664, 0.096008],
[0.4939998, -0.002244]]),
decimal=3)
assert_almost_equal(mlp.coefs_[1], np.array([[0.04706], [0.154089]]),
decimal=3)
assert_almost_equal(mlp.intercepts_[0],
np.array([0.098333, 0.09626]), decimal=3)
assert_almost_equal(mlp.intercepts_[1], np.array(0.9235), decimal=3)
# Testing output
# h1 = g(X1 * W_i1 + b11) = g(0.6 * 0.098 + 0.8 * 0.2956664 +
# 0.7 * 0.4939998 + 0.098333) = 0.677
# h2 = g(X2 * W_i2 + b12) = g(0.6 * 0.195756 + 0.8 * 0.096008 +
# 0.7 * -0.002244 + 0.09626) = 0.572
# o1 = h * W2 + b21 = 0.677 * 0.04706 +
# 0.572 * 0.154089 + 0.9235 = 1.043
# prob = sigmoid(o1) = 0.739
assert_almost_equal(mlp.predict_proba(X)[0, 1], 0.739, decimal=3)
import numpy as np
from sklearn.neural_network import MLPClassifier
os.chdir('../data')
# Load data
X = pd.read_csv('X.csv', header=None)
#with open('X.csv') as datafile:
# X = [[float(val) for val in line.split(',')] for line in datafile.read().split('\n') if line != '']
y = pd.read_csv('y.csv', header=None).transpose().values[0]
#with open('y.csv') as datafile:
# y = [float(line) for line in datafile.read().split('\n') if line != '']
Theta1 = pd.read_csv('Theta1.csv', header=None)
#with open('Theta1.csv') as datafile:
# Theta1 = [[float(val) for val in line.split(',')] for line in datafile.read().split('\n') if line != '']
Theta2 = pd.read_csv('Theta2.csv', header=None)
#with open('Theta2.csv') as datafile:
# Theta2 = [[float(val) for val in line.split(',')] for line in datafile.read().split('\n') if line != '']
clf = MLPClassifier(algorithm='l-bfgs', alpha=1e-5, hidden_layer_sizes=(25), random_state=1)
clf.fit(X, y)
#clf.n_outputs_ = 10
#clf.n_layers_ = 3
clf.out_activation_ = 'logistic'
clf.intercepts_ = [Theta1[0].values, Theta2[0].values]
clf.coefs_ = [Theta1.transpose()[1:].values, Theta2.transpose()[1:].values]
p = clf.predict(X)
diff = (p == y).mean()
print(diff)
def update(self):
if not World.executions:
raise RuntimeError('Cannot update. No executions left to work on.')
# Define workspace (data set of interest).
params = World.executions.pop(0)
World.params = params
t_time = time()
print('Data set %s %f' % (params['name'], t_time))
ds = params['dataset']()
X, y = ds.data, ds.target
if params['settings'].get('pc_decomposing', False):
X = decomposition.PCA(whiten=params['settings'].get(
'whiten', False),
random_state=0).fit_transform(X)
if params['settings']['plotting']:
plt.scatter(X[:, 0], X[:, 1], c=y, cmap=plt.cm.Spectral)
plt.tight_layout()
plt.savefig('report/ds-%s.png' % params['name'])
plt.close()
# Separate train (80%) and test data (20%).
X, X_test, y, y_test = model_selection.train_test_split(
X, y, train_size=.8, random_state=random_state)
# Set class attributes. We need this for the genetic algorithm.
World.X, World.X_test, World.y, World.y_test = X, X_test, y, y_test
# Build a regularly trained Neural Network once.
# We'll use it as base for our benchmarks.
mpl = MLPClassifier(**params['nn_params'])
print('Regular training ongoing...')
t = time()
mpl.fit(X, y)
print('Training complete (elapsed: %f s)' % (time() - t))
self.evaluate(mpl, label='NN')
best_i, best_model, best_score = -1, None, -np.inf
for i, search_params in enumerate(params['searches']):
trainer = art.agents.ResponderAgent(
search=art.searches.genetic.GeneticAlgorithm,
environment=self,
search_params=search_params)
# Ask agent to find a trained net for us.
print('Genetic training has started. Parameters: \n%s' %
search_params)
t = time()
training = trainer.act()
print(
'Evolution complete (%i cycles, %f s elapsed, '
'candidate utility: %f)' %
(trainer.search.cycle_, time() - t, trainer.utility(training)))
if (params['settings']['plotting']
and search_params.get('debug', False)):
# Plotting generations' utilities.
plt.plot(trainer.search.lowest_utility_,
color='blue',
linewidth=4,
label='Lowest')
plt.plot(trainer.search.average_utility_,
color='orange',
linewidth=4,
label='Average')
plt.plot(trainer.search.highest_utility_,
color='red',
linewidth=4,
label='Highest')
plt.legend()
plt.xlabel('generation')
plt.ylabel('utility')
plt.tight_layout()
plt.savefig('report/ut-%s-%i.png' % (params['name'], i))
plt.close()
# Let's cross our fingers! Build a Neural Network with the
# parameters selected by the evolutionary process.
gmpl = MLPClassifier(**params['nn_params'])
gmpl.coefs_ = training.coefs_
gmpl.intercepts_ = training.intercepts_
score = self.evaluate(gmpl, label='GA')
if score > best_score:
best_i, best_model, best_score = i, gmpl, score
gmpl.fit(X, y)
self.evaluate(gmpl, label='trained-gnn')
print('%s\'s report:\n'
'\tBest estimator id: %i\n'
'Score: %.2f\n'
'Total time elapsed: %f s\n'
'---\n' % (params['name'], best_i, best_score,
(time() - t_time)))
memory[index % memory_size,
0], memory[index % memory_size,
1] = observation[0], observation[1]
memory[index % memory_size, 2], memory[index % memory_size, 3], memory[
index % memory_size,
4] = action_0, observation_1[0], observation_1[1]
index += 1
observation = observation_1
# choose minimum batch
batch_index = np.random.choice(memory_size, size=mini_batch)
batch_memory = memory[batch_index, :]
e_r_l_new = batch_memory[:, 0:3]
e_r_l_old = batch_memory[:, 3:5]
q_target = np.zeros((mini_batch, 1))
counter_2 = 0
for element in e_r_l_old:
q_target[counter_2] = r_ + GAMMA * max(old_nn.predict([np.append(element, 0)]), \
old_nn.predict([np.append(element, 1)]), \
old_nn.predict([np.append(element, 2)]))
counter_2 += 1
q_target = q_target.ravel()
new_nn.fit(e_r_l_new, q_target)
if step % learning_inter == 0:
old_nn.coefs_ = new_nn.coefs_
old_nn.intercepts_ = new_nn.intercepts_
# reach the right top
if observation[0] == env.observation_space.high[0]:
break
clf = NN(hid_layers, activation='relu')
features, encodings, ((x_train, y_train), (x_val, y_val),
(x_test, y_test)) = process(all_data,
train_frac,
val_frac,
test_frac,
modify=True)
print(x_train.shape, y_train.shape)
clf.fit(x_train, y_train)
print("Simulated annealing:")
acc_anneal = []
test_anneal = []
clf.coefs_ = []
clf.intercepts_ = []
#anneal(clf, hid_layers, x_train, y_train) #uses simulated annealing to find the optimal weights
anneal = NNAnneal(clf, hid_layers, x_train, y_train)
([clf.coefs_, clf.intercepts_]), e = anneal.anneal()
print(accuracy_score(clf.predict(x_train), y_train))
print(accuracy_score(clf.predict(x_val), y_val))
print(accuracy_score(clf.predict(x_test), y_test))
# acc_anneal.append(accuracy_score(clf.predict(x_val), y_val))
# test_anneal.append(accuracy_score(clf.predict(x_test), y_test))
# print(acc_anneal)
# print(test_anneal)
print("Hill climbing:")
clf.coefs_ = []
def update(self):
if not World.executions:
raise RuntimeError('Cannot update. No executions left to work on.')
# Define workspace (data set of interest).
params = World.executions.pop(0)
World.params = params
t_time = time()
print('Data set %s %f' % (params['name'], t_time))
ds = params['dataset']()
X, y = ds.data, ds.target
if params['settings'].get('pc_decomposing', False):
X = decomposition.PCA(
whiten=params['settings'].get('whiten', False),
random_state=0).fit_transform(X)
if params['settings']['plotting']:
plt.scatter(X[:, 0], X[:, 1], c=y, cmap=plt.cm.Spectral)
plt.tight_layout()
plt.savefig('report/ds-%s.png' % params['name'])
plt.close()
# Separate train (80%) and test data (20%).
X, X_test, y, y_test = model_selection.train_test_split(
X, y, train_size=.8, random_state=random_state)
# Set class attributes. We need this for the genetic algorithm.
World.X, World.X_test, World.y, World.y_test = X, X_test, y, y_test
# Build a regularly trained Neural Network once.
# We'll use it as base for our benchmarks.
mpl = MLPClassifier(**params['nn_params'])
print('Regular training ongoing...')
t = time()
mpl.fit(X, y)
print('Training complete (elapsed: %f s)' % (time() - t))
self.evaluate(mpl, label='NN')
best_i, best_model, best_score = -1, None, -np.inf
for i, search_params in enumerate(params['searches']):
trainer = art.agents.ResponderAgent(
search=art.searches.genetic.GeneticAlgorithm,
environment=self,
search_params=search_params)
# Ask agent to find a trained net for us.
print('Genetic training has started. Parameters: \n%s'
% search_params)
t = time()
training = trainer.act()
print('Evolution complete (%i cycles, %f s elapsed, '
'candidate utility: %f)'
% (trainer.search.cycle_, time() - t,
trainer.utility(training)))
if (params['settings']['plotting'] and
search_params.get('debug', False)):
# Plotting generations' utilities.
plt.plot(trainer.search.lowest_utility_,
color='blue', linewidth=4, label='Lowest')
plt.plot(trainer.search.average_utility_,
color='orange', linewidth=4, label='Average')
plt.plot(trainer.search.highest_utility_,
color='red', linewidth=4, label='Highest')
plt.legend()
plt.xlabel('generation')
plt.ylabel('utility')
plt.tight_layout()
plt.savefig('report/ut-%s-%i.png' % (params['name'], i))
plt.close()
# Let's cross our fingers! Build a Neural Network with the
# parameters selected by the evolutionary process.
gmpl = MLPClassifier(**params['nn_params'])
gmpl.coefs_ = training.coefs_
gmpl.intercepts_ = training.intercepts_
score = self.evaluate(gmpl, label='GA')
if score > best_score:
best_i, best_model, best_score = i, gmpl, score
gmpl.fit(X, y)
self.evaluate(gmpl, label='trained-gnn')
print('%s\'s report:\n'
'\tBest estimator id: %i\n'
'Score: %.2f\n'
'Total time elapsed: %f s\n'
'---\n'
% (params['name'], best_i, best_score, (time() - t_time)))
