怎样健网站,上海计算机培训机构,wordpress怎么做图文数据库,本地wordpress安装遗传算法与深度学习实战——利用进化计算优化深度学习模型 0. 前言1. 利用进化计算优化深度学习模型2. 利用进化策略优化深度学习模型3. 利用差分计算优化深度学习模型相关链接 0. 前言
我们已经学习了使用进化策略 (Evolutionary Strategies, ES) 和差分进化 (Differential E… 遗传算法与深度学习实战——利用进化计算优化深度学习模型 0. 前言1. 利用进化计算优化深度学习模型2. 利用进化策略优化深度学习模型3. 利用差分计算优化深度学习模型相关链接 0. 前言
我们已经学习了使用进化策略 (Evolutionary Strategies, ES) 和差分进化 (Differential Evolution, DE) 调整超参数并取得了不错的结果。在本节中我们将应用 ES 和 DE 作为神经进化优化器并且继续使用圆圈或月亮数据集。
1. 利用进化计算优化深度学习模型
我们已经通过遗传算法优化简单深度学习 ( Deep learning, DL) 网络的权重/参数除了遗传算法之外我们还有更强大的进化方法比如进化策略 (Evolutionary Strategies, ES) 和差分进化 (Differential Evolution, DE)它们可能表现更好。接下来我们将研究这两种更高级的进化方法。 下图显示了使用 ES 和 DE 算法的运行结果并观察 DE 和 ES 如何演化在每个数据集的权重。可以看到ES不仅在解决困难数据集方面表现出色而且还有解决更复杂问题的潜力。 2. 利用进化策略优化深度学习模型
在代码中将遗传算法转换为进化策略 (Evolutionary Strategies, ES) 和差分进化 (Differential Evolution, DE) 算法。首先使用 ES 算法
import numpy as np
import sklearn
import sklearn.datasets
import sklearn.linear_model
import matplotlib.pyplot as plt
from IPython.display import clear_outputfrom deap import algorithms
from deap import base
from deap import benchmarks
from deap import creator
from deap import toolsimport randomnumber_samples 1000 #param {type:slider, min:100, max:1000, step:25}
difficulty 5 #param {type:slider, min:1, max:5, step:1}
problem classification #param [classification, blobs, gaussian quantiles, moons, circles]
number_features 2
number_classes 2
middle_layer 25 #param {type:slider, min:5, max:25, step:1}def load_data(problem): if problem classification:clusters 1 if difficulty 3 else 2informs 1 if difficulty 4 else 2data sklearn.datasets.make_classification(n_samples number_samples,n_featuresnumber_features, n_redundant0, class_sep1/difficulty,n_informativeinforms, n_clusters_per_classclusters)if problem blobs:data sklearn.datasets.make_blobs(n_samples number_samples,n_featuresnumber_features, centersnumber_classes,cluster_std difficulty)if problem gaussian quantiles:data sklearn.datasets.make_gaussian_quantiles(meanNone, covdifficulty,n_samplesnumber_samples,n_featuresnumber_features,n_classesnumber_classes,shuffleTrue,random_stateNone)if problem moons:data sklearn.datasets.make_moons(n_samples number_samples)if problem circles:data sklearn.datasets.make_circles(n_samples number_samples)return datadata load_data(problem)
X, Y data# Input Data
plt.figure(Input Data)
plt.scatter(X[:, 0], X[:, 1], cY, s40, cmapplt.cm.Spectral)def show_predictions(model, X, Y, name): display the labeled data X and a surface of prediction of model x_min, x_max X[:, 0].min() - 1, X[:, 0].max() 1y_min, y_max X[:, 1].min() - 1, X[:, 1].max() 1xx, yy np.meshgrid(np.arange(x_min, x_max, 0.01), np.arange(y_min, y_max, 0.01))X_temp np.c_[xx.flatten(), yy.flatten()]Z model.predict(X_temp)plt.figure(Predictions name)plt.contourf(xx, yy, Z.reshape(xx.shape), cmapplt.cm.Spectral)plt.ylabel(x2)plt.xlabel(x1)
plt.scatter(X[:, 0], X[:, 1],cY, s40, cmapplt.cm.Spectral)clf sklearn.linear_model.LogisticRegressionCV()
clf.fit(X, Y)show_predictions(clf, X, Y, Logistic regression)LR_predictions clf.predict(X)
print(Logistic Regression accuracy : , np.sum(LR_predictions Y) / Y.shape[0])def sigmoid(x):return 1.0 / (1.0 np.exp(-x)) ## Neural Network
class Neural_Network:def __init__(self, n_in, n_hidden, n_out):# Network dimensionsself.n_x n_inself.n_h n_hiddenself.n_y n_out# Parameters initializationself.W1 np.random.randn(self.n_h, self.n_x) * 0.01self.b1 np.zeros((self.n_h, 1))self.W2 np.random.randn(self.n_y, self.n_h) * 0.01self.b2 np.zeros((self.n_y, 1))self.parameters [self.W1, self.b1, self.W2, self.b2]def forward(self, X): Forward computation self.Z1 self.W1.dot(X.T) self.b1self.A1 np.tanh(self.Z1)self.Z2 self.W2.dot(self.A1) self.b2self.A2 sigmoid(self.Z2)def set_parameters(self, individual):Sets model parameters idx 0for p in self.parameters: size p.sizesh p.shapet individual[idx:idxsize]t np.array(t)t np.reshape(t, sh)p - pp tidx sizedef predict(self, X): Compute predictions with just a forward pass self.forward(X)return np.round(self.A2).astype(np.int)nn Neural_Network(2, middle_layer, 1)
number_of_genes sum([p.size for p in nn.parameters])
print(number_of_genes)individual np.ones(number_of_genes)
nn.set_parameters(individual)
print(nn.parameters)IND_SIZE number_of_genes
MIN_VALUE -1000
MAX_VALUE 1000
MIN_STRATEGY 0.5
MAX_STRATEGY 5CXPB .6
MUTPB .3creator.create(FitnessMin, base.Fitness, weights(-1.0,))
creator.create(Individual, list, typecoded, fitnesscreator.FitnessMin, strategyNone)
creator.create(Strategy, list, typecoded)def generateES(icls, scls, size, imin, imax, smin, smax): ind icls(random.uniform(imin, imax) for _ in range(size)) ind.strategy scls(random.uniform(smin, smax) for _ in range(size)) return inddef checkStrategy(minstrategy):def decorator(func):def wrappper(*args, **kargs):children func(*args, **kargs)for child in children:for i, s in enumerate(child.strategy):if s minstrategy:child.strategy[i] minstrategyreturn childrenreturn wrappper
return decoratordef custom_blend(ind1, ind2, alpha): for i, (x1, s1, x2, s2) in enumerate(zip(ind1, ind1.strategy,ind2, ind2.strategy)):# Blend the valuesgamma (1. 2. * alpha) * random.random() - alphaind1[i] (1. - gamma) * x1 gamma * x2ind2[i] gamma * x1 (1. - gamma) * x2# Blend the strategiesgamma (1. 2. * alpha) * random.random() - alphaind1.strategy[i] (1. - gamma) * s1 gamma * s2ind2.strategy[i] gamma * s1 (1. - gamma) * s2
return ind1, ind2toolbox base.Toolbox()
toolbox.register(individual, generateES, creator.Individual, creator.Strategy,IND_SIZE, MIN_VALUE, MAX_VALUE, MIN_STRATEGY, MAX_STRATEGY)
toolbox.register(population, tools.initRepeat, list, toolbox.individual)
toolbox.register(mate, custom_blend, alpha0.5)
toolbox.register(mutate, tools.mutESLogNormal, c1.0, indpb0.06)
toolbox.register(select, tools.selTournament, tournsize5)toolbox.decorate(mate, checkStrategy(MIN_STRATEGY))
toolbox.decorate(mutate, checkStrategy(MIN_STRATEGY))nn Neural_Network(2, middle_layer, 1)
nn.set_parameters(individual)
print(nn.parameters)show_predictions(nn, X, Y, Neural Network)nn_predictions nn.predict(X)
print(Neural Network accuracy : , np.sum(nn_predictions Y) / Y.shape[0])def evaluate(individual): nn.set_parameters(individual)nn_predictions nn.predict(X)return 1/np.sum(nn_predictions Y) / Y.shape[0], toolbox.register(evaluate, evaluate)MU 340 #param {type:slider, min:5, max:1000, step:5}
LAMBDA 1000 #param {type:slider, min:5, max:1000, step:5}
NGEN 100 #param {type:slider, min:100, max:1000, step:10}
RGEN 10 #param {type:slider, min:1, max:100, step:1}
CXPB .6
MUTPB .3random.seed(64)pop toolbox.population(nMU)
hof tools.HallOfFame(1)
stats tools.Statistics(lambda ind: ind.fitness.values)
stats.register(avg, np.mean)
stats.register(std, np.std)
stats.register(min, np.min)
stats.register(max, np.max)best None
history []for g in range(NGEN):pop, logbook algorithms.eaMuCommaLambda(pop, toolbox, muMU, lambda_LAMBDA, cxpbCXPB, mutpbMUTPB, ngenRGEN, statsstats, halloffamehof, verboseFalse)best hof[0] clear_output()print(fGen ({(g1)*RGEN}))show_predictions(nn, X, Y, Neural Network) nn_predictions nn.predict(X)print(Current Neural Network accuracy : , np.sum(nn_predictions Y) / Y.shape[0])plt.show()nn.set_parameters(best)show_predictions(nn, X, Y, Best Neural Network)plt.show()nn_predictions nn.predict(X)fitness np.sum(nn_predictions Y) / Y.shape[0]print(Best Neural Network accuracy : , fitness)if fitness .99: #stop conditionbreak3. 利用差分计算优化深度学习模型
接下来应用 DE 作为神经进化优化器
import numpy as np
import sklearn
import sklearn.datasets
import sklearn.linear_model
import matplotlib.pyplot as plt
from IPython.display import clear_outputfrom deap import algorithms
from deap import base
from deap import benchmarks
from deap import creator
from deap import toolsimport random
import array
import timenumber_samples 1000 #param {type:slider, min:100, max:1000, step:25}
difficulty 5 #param {type:slider, min:1, max:5, step:1}
problem classification #param [classification, blobs, gaussian quantiles, moons, circles]
number_features 2
number_classes 2
middle_layer 25 #param {type:slider, min:5, max:25, step:1}def load_data(problem): if problem classification:clusters 1 if difficulty 3 else 2informs 1 if difficulty 4 else 2data sklearn.datasets.make_classification(n_samples number_samples,n_featuresnumber_features, n_redundant0, class_sep1/difficulty,n_informativeinforms, n_clusters_per_classclusters)if problem blobs:data sklearn.datasets.make_blobs(n_samples number_samples,n_featuresnumber_features, centersnumber_classes,cluster_std difficulty)if problem gaussian quantiles:data sklearn.datasets.make_gaussian_quantiles(meanNone, covdifficulty,n_samplesnumber_samples,n_featuresnumber_features,n_classesnumber_classes,shuffleTrue,random_stateNone)if problem moons:data sklearn.datasets.make_moons(n_samples number_samples)if problem circles:data sklearn.datasets.make_circles(n_samples number_samples)return datadata load_data(problem)
X, Y data# Input Data
plt.figure(Input Data)
plt.scatter(X[:, 0], X[:, 1], cY, s40, cmapplt.cm.Spectral)def show_predictions(model, X, Y, name): display the labeled data X and a surface of prediction of model x_min, x_max X[:, 0].min() - 1, X[:, 0].max() 1y_min, y_max X[:, 1].min() - 1, X[:, 1].max() 1xx, yy np.meshgrid(np.arange(x_min, x_max, 0.01), np.arange(y_min, y_max, 0.01))X_temp np.c_[xx.flatten(), yy.flatten()]Z model.predict(X_temp)plt.figure(Predictions name)plt.contourf(xx, yy, Z.reshape(xx.shape), cmapplt.cm.Spectral)plt.ylabel(x2)plt.xlabel(x1)
plt.scatter(X[:, 0], X[:, 1],cY, s40, cmapplt.cm.Spectral)clf sklearn.linear_model.LogisticRegressionCV()
clf.fit(X, Y)show_predictions(clf, X, Y, Logistic regression)LR_predictions clf.predict(X)
print(Logistic Regression accuracy : , np.sum(LR_predictions Y) / Y.shape[0])def sigmoid(x):return 1.0 / (1.0 np.exp(-x)) ## Neural Network
class Neural_Network:def __init__(self, n_in, n_hidden, n_out):# Network dimensionsself.n_x n_inself.n_h n_hiddenself.n_y n_out# Parameters initializationself.W1 np.random.randn(self.n_h, self.n_x) * 0.01self.b1 np.zeros((self.n_h, 1))self.W2 np.random.randn(self.n_y, self.n_h) * 0.01self.b2 np.zeros((self.n_y, 1))self.parameters [self.W1, self.b1, self.W2, self.b2]def forward(self, X): Forward computation self.Z1 self.W1.dot(X.T) self.b1self.A1 np.tanh(self.Z1)self.Z2 self.W2.dot(self.A1) self.b2self.A2 sigmoid(self.Z2)def set_parameters(self, individual):Sets model parameters idx 0for p in self.parameters: size p.sizesh p.shapet individual[idx:idxsize]t np.array(t)t np.reshape(t, sh)p - pp tidx sizedef predict(self, X): Compute predictions with just a forward pass self.forward(X)return np.round(self.A2).astype(np.int)nn Neural_Network(2, middle_layer, 1)
number_of_genes sum([p.size for p in nn.parameters])
print(number_of_genes)individual np.ones(number_of_genes)
nn.set_parameters(individual)
print(nn.parameters)NDIM number_of_genes
CR 0.25
F_ 1creator.create(FitnessMin, base.Fitness, weights(-1.0,))
creator.create(Individual, array.array, typecoded, fitnesscreator.FitnessMin)toolbox base.Toolbox()
toolbox.register(attr_float, random.uniform, -1, 1)
toolbox.register(individual, tools.initRepeat, creator.Individual, toolbox.attr_float, NDIM)
toolbox.register(population, tools.initRepeat, list, toolbox.individual)
toolbox.register(select, tools.selRandom, k3)nn Neural_Network(2, middle_layer, 1)
nn.set_parameters(individual)
print(nn.parameters)show_predictions(nn, X, Y, Neural Network)nn_predictions nn.predict(X)
print(Neural Network accuracy : , np.sum(nn_predictions Y) / Y.shape[0])def evaluate(individual): nn.set_parameters(individual)nn_predictions nn.predict(X)return 1/np.sum(nn_predictions Y) / Y.shape[0], toolbox.register(evaluate, evaluate)MU 340 #param {type:slider, min:5, max:1000, step:5}
NGEN 1000 #param {type:slider, min:100, max:1000, step:10}
RGEN 10 #param {type:slider, min:1, max:10, step:1}random.seed(64)pop toolbox.population(nMU)
hof tools.HallOfFame(1)
stats tools.Statistics(lambda ind: ind.fitness.values)
stats.register(avg, np.mean)
stats.register(std, np.std)
stats.register(min, np.min)
stats.register(max, np.max)logbook tools.Logbook()
logbook.header gen, evals, std, min, avg, maxbest None
history []# Evaluate the individuals
fitnesses toolbox.map(toolbox.evaluate, pop)
for ind, fit in zip(pop, fitnesses):ind.fitness.values fitrecord stats.compile(pop)
logbook.record(gen0, evalslen(pop), **record)
print(logbook.stream)
for g in range(1, NGEN):for k, agent in enumerate(pop):a,b,c toolbox.select(pop)y toolbox.clone(agent)index random.randrange(NDIM)for i, value in enumerate(agent):if i index or random.random() CR:y[i] a[i] F_*(b[i]-c[i])y.fitness.values toolbox.evaluate(y)if y.fitness agent.fitness:pop[k] yhof.update(pop) record stats.compile(pop) best hof[0]if ((g1) % RGEN) 0:clear_output()print(fGen ({(g1)}))show_predictions(nn, X, Y, Neural Network) nn_predictions nn.predict(X)print(Current Neural Network accuracy : , np.sum(nn_predictions Y) / Y.shape[0])plt.show()nn.set_parameters(best)show_predictions(nn, X, Y, Best Neural Network)plt.show()nn_predictions nn.predict(X)fitness np.sum(nn_predictions Y) / Y.shape[0]print(Best Neural Network accuracy : , fitness)if fitness .99: #stop conditionbreaktime.sleep(1)对于本节所用的样本数据集简单的遗传算法方法通常表现最佳DE 明显是不太理想的选择而 ES 具有一定的潜力。可以通过完成以下问题进一步了解不同神经优化算法的应用场景
调整超参数观察它们对 DE 或 ES 的影响。调整特定的进化方法超参数——ES 的最小和最大策略或 DE 的 pmin/pmax 和 smin/smax
相关链接
遗传算法与深度学习实战1——进化深度学习 遗传算法与深度学习实战4——遗传算法Genetic Algorithm详解与实现 遗传算法与深度学习实战14——进化策略详解与实现 遗传算法与深度学习实战15——差分进化详解与实现 遗传算法与深度学习实战16——神经网络超参数优化 遗传算法与深度学习实战19——使用粒子群优化自动超参数优化 遗传算法与深度学习实战20——使用进化策略自动超参数优化 遗传算法与深度学习实战21——使用差分搜索自动超参数优化 遗传算法与深度学习实战22——使用Numpy构建神经网络 遗传算法与深度学习实战23——利用遗传算法优化深度学习模型
