cur.Ystd = Ystd_cur # As above but for the labels #Lmean_cur = L_cur.mean() #Ln_cur = L_cur - Lmean_cur #Lstd_cur = Ln_cur.std() #Ln_cur /= Lstd_cur #Ltestn_cur = Ltest_cur - Lmean_cur #Ltestn_cur /= Lstd_cur cur.X=None cur.Y = {'Y':Yn_cur} cur.Ytestn = {'Ytest':Ytestn_cur} cur.Ltest = {'Ltest':Ltest_cur} fname_cur = fname + '_L' + str(i) cur.training(model_num_inducing, model_num_iterations, model_init_iterations, fname_cur, save_model, economy_save) mm.append(cur) ss = []; sstest = []; for i in range(len(Lunique)): for j in range(len(Lunique)): ss = mm[i].SAMObject.familiarity(mm[j].Y['Y']) print('Familiarity of model ' + participantList[i] + ' given label: ' + participantList[j] + ' using training data is: ' + str(ss)) print("") print("") print("") for i in range(len(Lunique)): for j in range(len(Lunique)): sstest = mm[i].SAMObject.familiarity(mm[j].Ytestn['Ytest'])
# Specification of model type and training parameters model_type = 'mrd' model_num_inducing = 30 model_num_iterations = 150 model_init_iterations = 400 fname = modelPath + '/models/' + 'mActions_' + model_type + '_exp' + str(experiment_number) #+ '.pickle' # Enable to save the model and visualise GP nearest neighbour matching save_model=True visualise_output=True # Reading face data, preparation of data and training of the model mySAMpy.readData(root_data_dir, participant_index, pose_index) mySAMpy.prepareData(model_type, Ntr, pose_selection) mySAMpy.training(model_num_inducing, model_num_iterations, model_init_iterations, fname, save_model) while( not(yarp.Network.isConnected("/speechInteraction/behaviour:o","/sam/face/interaction:i")) ): print "Waiting for connection with behaviour port..." pass # This is for visualising the mapping of the test face back to the internal memory if visualise_output: ax = mySAMpy.SAMObject.visualise() visualiseInfo=dict() visualiseInfo['ax']=ax ytmp = mySAMpy.SAMObject.recall(0) ytmp = numpy.reshape(ytmp,(mySAMpy.imgHeightNew,mySAMpy.imgWidthNew)) fig_nn = pb.figure() pb.title('Training NN')
model_init_iterations = 400 fname = modelPath + '/models/' + 'mActions_' + model_type + '_exp' + str( experiment_number) #+ '.pickle' # Enable to save the model and visualise GP nearest neighbour matching save_model = True economy_save = True # ATTENTION!! This is still BETA!! visualise_output = True # Reading face data, preparation of data and training of the model mySAMpy.readData(root_data_dir, participant_index, pose_index) mySAMpy.prepareData(model_type, Ntr, pose_selection, randSeed=experiment_number) mySAMpy.training(model_num_inducing, model_num_iterations, model_init_iterations, fname, save_model, economy_save) if yarpRunning: while (not (yarp.Network.isConnected("/speechInteraction/behaviour:o", "/sam/face/interaction:i"))): print "Waiting for connection with behaviour port..." pass # This is for visualising the mapping of the test face back to the internal memory if visualise_output: ax = mySAMpy.SAMObject.visualise() visualiseInfo = dict() visualiseInfo['ax'] = ax ytmp = mySAMpy.SAMObject.recall(0) ytmp = numpy.reshape(ytmp, (mySAMpy.imgHeightNew, mySAMpy.imgWidthNew)) fig_nn = pb.figure()
#Lmean_cur = L_cur.mean() #Ln_cur = L_cur - Lmean_cur #Lstd_cur = Ln_cur.std() #Ln_cur /= Lstd_cur #Ltestn_cur = Ltest_cur - Lmean_cur #Ltestn_cur /= Lstd_cur cur.X = None cur.Y = {'Y':Yn_cur} cur.Ytestn = {'Ytest':Ytestn_cur} cur.Ltest = {'Ltest':Ltest_cur} print 'training data' + str(cur.Y['Y'].shape) print 'testing data' + str(cur.Ytestn['Ytest'].shape) fname_cur = fname + '__L' + str(i) cur.training(model_num_inducing, model_num_iterations, model_init_iterations, fname_cur, save_model, economy_save, keepIfPresent=False, kernelStr=kernelString) mm.append(cur) ss = []; sstest = []; print #Ntest = 10 print Ntest result = np.zeros([len(participantList),Ntest,len(participantList)]) responseIdx = np.zeros([result.shape[0],result.shape[1]]) responseVal = np.zeros([result.shape[0],result.shape[1]]) confusionMatrix = np.zeros([result.shape[0],result.shape[0]]) for i in range(result.shape[0]): print print('Participant ' + str(i) + ': ' + participantList[i])
cur.Ystd = Ystd_cur # As above but for the labels #Lmean_cur = L_cur.mean() #Ln_cur = L_cur - Lmean_cur #Lstd_cur = Ln_cur.std() #Ln_cur /= Lstd_cur #Ltestn_cur = Ltest_cur - Lmean_cur #Ltestn_cur /= Lstd_cur cur.X = None cur.Y = {'Y': Yn_cur} cur.Ytestn = {'Ytest': Ytestn_cur} cur.Ltest = {'Ltest': Ltest_cur} fname_cur = fname + '_L' + str(i) cur.training(model_num_inducing, model_num_iterations, model_init_iterations, fname_cur, save_model, economy_save) mm.append(cur) ss = [] sstest = [] for i in range(len(Lunique)): for j in range(len(Lunique)): ss = mm[i].SAMObject.familiarity(mm[j].Y['Y']) print('Familiarity of model ' + participantList[i] + ' given label: ' + participantList[j] + ' using training data is: ' + str(ss)) print("") print("") print("") for i in range(len(Lunique)): for j in range(len(Lunique)):