from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter from keras.datasets import mnist from tqdm import tqdm from Network.Net import WTA, SpikeMonitor, seed from Network.Tools import plot_NonSp, plot_AvrInp, plot_MissClass, plot_ConfMtx from brian2.units import * import matplotlib.pyplot as plt import numpy as np import yaml import sys def get_Spikes(X_data:np.ndarray, init_params:dict, Net_params:dict, presen_stg:str='Train', Run_trial:bool=False): seed(init_params['Random_Seed']) if presen_stg == 'Train': dir_data = 'Activity/Train/' state_title = 'MAPPING' elif presen_stg == 'Test': dir_data = 'Activity/Test/' state_title = 'VALIDATING' if init_params['Load_Temp'] == True: addr_load = 'Temp/' else: addr_load = 'Trained_Models/' if Run_trial: print("================== # "+ state_title +" MODEL # ==================") Mdl = WTA(Net_setup=Net_params) Mdl.Init_State() X_pre = Mdl.preProcess(X_data=X_data, preInp=init_params['Gabor_filter']) Train_Sp, Input_Sp = [], [] for idx in tqdm(range(len(X_pre)), desc='Validating'): Mdl.net.restore(init_params['Filename'], filename=addr_load + init_params['Filename'] + '.b2') # Rate Monitor for Counting Spikes mon = SpikeMonitor(Mdl.net['Exc'], name='CountSp') Mdl.net.add(mon) if presen_stg == 'Test': monInp = SpikeMonitor(Mdl.net['Input'], name='Count_InpSp') Mdl.net.add(monInp) Mdl.Norm_SynW(Norm_w=True) Mdl.RunModel(X_single=X_pre[idx], preInp=init_params['Gabor_filter'], norm=init_params['Norm'], phase='Stimulus') Train_Sp.append(np.array(mon.count, dtype=np.int8)) if presen_stg == 'Test': Input_Sp.append(np.array(monInp.count, dtype=np.int8)) Mdl.RunModel(phase='Resting') Mdl.net.remove(Mdl.net['CountSp']) if presen_stg == 'Test': Mdl.net.remove(Mdl.net['Count_InpSp']) np.save(dir_data + init_params['Filename'] + '_' + str(len(X_data)), Train_Sp) if presen_stg == 'Test': np.save('Activity/Test/Input/Poisson_Count_' + str(len(X_data)), Input_Sp) else: Train_Sp = np.load(dir_data + init_params['Filename'] + '_' + str(len(X_data)) + '.npy') print('Loaded Excitatory Spike Data Shape: ' + str(np.array(Train_Sp).shape)) if presen_stg == 'Test': Input_Sp = np.load('Activity/Test/Input/Poisson_Count_' + str(len(X_data)) + '.npy') print('Loaded Input Count Data Shape: ' + str(np.array(Input_Sp).shape)) if presen_stg == 'Test': return Train_Sp, Input_Sp else: return Train_Sp def assign_Class(data, Y): Y = np.array(Y) data_test = np.array(data) assignments = np.ones(len(data_test[0,:])) * -1 # initialize them as not assigned input_nums = np.asarray(Y[:len(data)]) maximum_rate = [0] * len(data_test[0,:]) for j in range(10): num_inputs = len(np.where(input_nums == j)[0]) if num_inputs > 0: # Sum found index MNIST value (row) to get the average of spikes (col) per MNIST input rate = np.sum(data_test[input_nums == j], axis = 0) / num_inputs for i in range(len(data_test[0,:])): if rate[i] > maximum_rate[i]: maximum_rate[i] = rate[i] assignments[i] = j return assignments def Calculate_Correct(data, Y, Class_Map): correct, result, correct_class_idx = 0, [], [] data = np.array(data) Class_Map = np.array(Class_Map) for idx in range(len(data)): max_val_idx = np.argmax(data[idx,:]) assign_val = int(Class_Map[max_val_idx]) result.append(assign_val) if assign_val == Y[idx]: correct += 1 correct_class_idx.append(idx) return correct, np.array(result), correct_class_idx def Str2bool(Val_arg): if Val_arg == "True": return True elif Val_arg == "False": return False if __name__ == "__main__": # ==================== Argument Initialization ======================== parser = ArgumentParser(formatter_class=ArgumentDefaultsHelpFormatter) parser.add_argument("-s", "--seed", default=0, type=int, help="Random Seed Initialization") parser.add_argument("-f", "--filename", default="default", type=str, help="Filename of the Model to be saved") parser.add_argument("-t", "--temp", default=False, type=Str2bool, help="Select to load the brian2 file from the temp directory") parser.add_argument("-gb", "--gabor", default=True, type=Str2bool, help="Preprocess Input data with Gabor Filter") parser.add_argument("-n", "--norm", default=True, type=Str2bool, help="Applied Input Normalization after Gabor Filter") parser.add_argument("-m", "--train_dt", default=1000, type=int, help="Length of dataset to train our model") parser.add_argument("-d", "--test_dt", default=10000, type=int, help="Length of dataset to test our model") parser.add_argument("-p", "--plot", default=False, type=Str2bool, help="Show the result plots after running the Validation process") parser.add_argument("-rm", "--run_train", default=True, type=Str2bool, help="Run a single presentation of the train dataset") parser.add_argument("-r", "--run_test", default=True, type=Str2bool, help="Run a single presentation of the test dataset") args = vars(parser.parse_args()) # =========================== Parameters ============================== with open('Network/params.yml', 'r') as file: net = yaml.safe_load(file) file.close() Validate_params = { 'Random_Seed':args['seed'], 'Filename':args['filename'], 'Load_Temp':args['temp'], 'Gabor_filter':args['gabor'], 'Norm':args['norm'], 'Train_dt':args['train_dt'], 'Test_dt':args['test_dt'], 'Run_train':args['run_train'], 'Run_test':args['run_test'] } Net_init = { 'Neurons':net['Net'][0], 'Learning_Rule':net['Net'][1], 'Nearest_Neighbor':net['Net'][2], 'Pre_Offset':net['Net'][3], 'Run_test':net['Validate'][0], 'Monitors':net['Validate'][1] } # ====================== Load MNIST Dataset ========================== (X_train, y_train), (X_test, y_test) = mnist.load_data() X_train = X_train / 4. X_test = X_test / 4. # ================== Get Train Activity Data ========================== train_dt = Validate_params['Train_dt'] Run_train = Validate_params['Run_train'] train_data = get_Spikes(X_data=X_train[:train_dt], init_params=Validate_params, Net_params=Net_init, presen_stg='Train', Run_trial=Run_train) # =================== Assign Labels to Data =========================== Inp_map = np.array(assign_Class(data=train_data, Y=y_train)) if np.sum(Inp_map) <= 0: print('No Spikes in Excitatory Layer!!') sys.exit(0) # =================== Get Test Activity Data =========================== test_dt = Validate_params['Test_dt'] Run_test = Validate_params['Run_test'] test_data, Inp_test = get_Spikes(X_data=X_test[:test_dt], init_params=Validate_params, Net_params=Net_init, presen_stg='Test', Run_trial=Run_test) # =================== Count Img with No Spike =========================== Sp_train = np.sum(train_data, axis=1) Sp_test = np.sum(test_data, axis=1) Test_sum = np.array(np.sum(Inp_test, axis=1)) NonSp_train = [idx for idx, Zero in enumerate(Sp_train) if Zero == 0] NonSp_test = [idx for idx, Zero in enumerate(Sp_test) if Zero == 0] TrueSp_test = [idx for idx, Zero in enumerate(Sp_test) if Zero != 0] # ======================= Print Accuracy Report ========================= correct, result, Class_idx = Calculate_Correct(data=test_data, Y=y_test, Class_Map=Inp_map) accuracy_r = (correct/test_dt) * 100 miss_arg = [idx for idx, Img_label in enumerate(result) if Img_label != y_test[idx]] True_miss_arg = np.setdiff1d(miss_arg, NonSp_test) np.save('Results/Accuracy/Res_' + Validate_params['Filename'], [accuracy_r, train_dt, (len(NonSp_train)/train_dt)*100, (len(NonSp_test)/test_dt)*100]) print('=============== ' + Validate_params['Filename'] + ' ===============') print('-----Excitatory Neuronal Layer Map-----') print(Inp_map) print('Correct Classified: ' + str(correct)) print('Missclassified: ' + str(test_dt - correct)) print('Accuracy: ' + str(accuracy_r) + ' %') print('No Spikes Train Pres: ' + str((len(NonSp_train)/train_dt)*100) + ' %') print('No Spikes Test Eval: ' + str((len(NonSp_test)/test_dt)*100) + ' %') # ==================== Plots of Network Behavior ====================== cycle_plots = args['plot'] plot_NonSp(Label_data=y_test[NonSp_test], dataset_type='Test') plot_AvrInp(Sp_Inp=Test_sum, Y_data=y_test, NonSp_idx=NonSp_test, Miss_idx=miss_arg, Correct_idx=Class_idx) plot_MissClass(y_arr=np.bincount(result[True_miss_arg])) plot_ConfMtx(Sp_pred=result, Label_true=y_test, TrueSp_idx=TrueSp_test) # it_lim = 25000 # acc_it = [np.load('Results/Accuracy/Res_Temp_It_' + str(num) + '.npy')[0] for num in np.arange(1000, it_lim+1000, 1000)] # x_arr = np.arange(1, len(acc_it)+1, 1) # plt.figure(figsize=(8,6)) # plt.plot(x_arr*1000, acc_it, color='purple', marker='o') # plt.ylabel('Accuracy [%]') # plt.xlabel('Iteration') # plt.xlim(0, it_lim+1000) # plt.ylim(40, 100) # plt.legend(['pair-wise STDP']) # plt.grid(True) # plt.tight_layout() plt.show(block=cycle_plots)