def worker(input_worker): """Explanations""" #Global variables: global numInput,numOutput,numHidden global dim_hidden_output, dim_hidden_output global sigma global env #Local: seed = int(input_worker[0]) p = input_worker[1] env.seed(seed) np.random.seed(seed) #Neural Networks: NN = NeuralNetwork(numInput,numHidden,numOutput, VBN_dict) NN.wi=p[0] NN.wo=p[1] #distortions epsilon = np.random.multivariate_normal(np.zeros(dim_hidden_output+dim_input_hidden),np.identity(dim_hidden_output+dim_input_hidden)) epsilon_wo= epsilon[0:dim_hidden_output].reshape((numHidden,numOutput)) epsilon_wi= epsilon[dim_hidden_output:dim_hidden_output+dim_input_hidden].reshape((numInput,numHidden)) #epsilon_wo = np.random.multivariate_normal([0 for x in range(dim_hidden_output)],np.identity(dim_hidden_output)).reshape((numHidden,numOutput)) #epsilon_wi = np.random.multivariate_normal([0 for x in range(dim_input_hidden)],np.identity(dim_input_hidden)).reshape((numInput,numHidden)) #parameters update NN.wo=NN.wo+epsilon_wo*sigma #remark:we should merge the two, and reshape the matrix NN.wi=NN.wi+epsilon_wi*sigma #initial_observation=env.reset() reward_worker=episodeRoute(NN,env,initial_observation,steps=250) return(reward_worker,epsilon_wi,epsilon_wo)
def worker_train_VBN(input_worker_VBN): """Explanations""" #Global variables: global numInput,numOutput,numHidden global dim_hidden_output, dim_hidden_output global env #Local: seed=int(input_worker_VBN[0]) p = input_worker_VBN[1] env.seed(seed) #np.random.seed(seed) VBN_dict = {} VBN_dict['mu_i']=0 VBN_dict['var_i']=0 VBN_dict['mu_h']=0 VBN_dict['var_h']=0 VBN_dict['mu_o']=0 VBN_dict['var_o']=0 #Neural Networks: NN = NeuralNetwork(numInput,numHidden,numOutput, VBN_dict) NN.wi=p[0] NN.wo=p[1] steps=250 ai = env.reset() num_step=steps for j in range(steps): ao = NN.feedForward(ai) #to transfer to the main # question: how many worker for this ? sum_zi=[0.] * numInput sum_zh=[0.] * numHidden sum_zo=[0.] *numOutput sum_zi2=[0.] * numInput sum_zh2=[0.] * numHidden sum_zo2=[0.] *numOutput sum_zi=[sum(x) for x in zip(sum_zi, NN.zi)] ### VERY WEIRD !! ALWAYS EQUAL TO 1 ? sum_zh=[sum(x) for x in zip(sum_zh, NN.zh)] sum_zo=[sum(x) for x in zip(sum_zo, NN.zo)] sum_zi2=[sum(x) for x in zip(sum_zi2, square(NN.zi))] sum_zh2=[sum(x) for x in zip(sum_zh2, square(NN.zh))] sum_zo2=[sum(x) for x in zip(sum_zo2, square(NN.zo))] ''' sum_zi=map(add, sum_zi, NN.zi) sum_zh=map(add, sum_zh, NN.zh) sum_zo=map(add, sum_zo, NN.zo) sum_zi=map(add, sum_zi2, square(NN.zi)) sum_zh=map(add, sum_zh2, square(NN.zh)) sum_zo=map(add, sum_zo2, square(NN.zo)) ''' action=np.argmax(ao) ai, reward, done, info = env.step(action) if done: break num_step=j return(sum_zi,sum_zh,sum_zo,sum_zi2,sum_zh2,sum_zo2,num_step)