CNTK_201B_CIFAR-10_ImageHandsOn の改造

''' CNTK_201B_CIFAR-10_ImageHandsOn_nishi.py オリジナルの CNTK_201B_CIFAR-10_ImageHandsOn をもとに データ入力部分を、 keras.dataset.cifar10 を使ってみます。 Created on 2018/06/02 @author: nishi ''' from IPython.display import Image import matplotlib.pyplot as plt import math import numpy as np import os import PIL import sys import cntk as C from scipy.misc import toimage from keras.datasets import cifar10 from sklearn.utils import shuffle ''' keras.dataset.cifar10 ライブラリを使って CIFAR-10データセットをロード cifar10.load_data() は、一度データをダウンロードすると、 どこかにキャッシュが残っていて、2回目からは、ダウンロードが省略される様です。 ''' (X_train, Y_train), (X_test, Y_test) = cifar10.load_data() Y_train = np.eye(10)[Y_train.astype(int)] # 1-of-k表現に変換 Y_test = np.eye(10)[Y_test.astype(int)] # 1-of-k表現に変換 X_train=X_train.astype(np.float32) X_test=X_test.astype(np.float32) Y_train=Y_train.astype(np.float32) Y_test=Y_test.astype(np.float32) print(X_train.shape, Y_train.shape) print(X_test.shape, Y_test.shape) # X_train.shape (50000,32,32,3) -> (50000,3,32,32) に変換が必要です。 X_train= np.rollaxis(X_train,3,1) X_test= np.rollaxis(X_test,3,1) print('X_train.shape=',X_train.shape) print('X_test.shape=',X_test.shape) #In[9]: print('In[9]:') def create_basic_model(input, out_dims): with C.layers.default_options(init=C.glorot_uniform(), activation=C.relu): net = C.layers.Convolution((5,5), 32, pad=True)(input) net = C.layers.MaxPooling((3,3), strides=(2,2))(net) net = C.layers.Convolution((5,5), 32, pad=True)(net) net = C.layers.MaxPooling((3,3), strides=(2,2))(net) net = C.layers.Convolution((5,5), 64, pad=True)(net) net = C.layers.MaxPooling((3,3), strides=(2,2))(net) net = C.layers.Dense(64)(net) net = C.layers.Dense(out_dims, activation=None)(net) return net #In[10]: print('In[10]:') # model dimensions image_height = 32 image_width = 32 num_channels = 3 num_classes = 10 #In[12]: print('In[12]:') # # Train and evaluate the network. # def train_and_evaluate(x_train,y_train,x_test,y_test, max_epochs, model_func): # Input variables denoting the features and label data input_var = C.input_variable((num_channels, image_height, image_width)) label_var = C.input_variable((num_classes)) # Normalize the input feature_scale = 1.0 / 256.0 input_var_norm = C.element_times(feature_scale, input_var) # apply model to input z = model_func(input_var_norm, out_dims=10) # # Training action # # loss and metric ce = C.cross_entropy_with_softmax(z, label_var) pe = C.classification_error(z, label_var) # training config epoch_size = 50000 minibatch_size = 64 n_batches = len(x_train) // minibatch_size # Set training parameters lr_per_minibatch = C.learning_parameter_schedule([0.01]*10 + [0.003]*10 + [0.001], epoch_size = epoch_size) momentums = C.momentum_schedule(0.9, minibatch_size = minibatch_size) l2_reg_weight = 0.001 # trainer object learner = C.momentum_sgd(z.parameters, lr = lr_per_minibatch, momentum = momentums, l2_regularization_weight=l2_reg_weight) progress_printer = C.logging.ProgressPrinter(tag='Training', num_epochs=max_epochs) trainer = C.Trainer(z, (ce, pe), [learner], [progress_printer]) # define mapping from reader streams to network inputs #input_map = { # input_var: reader_train.streams.features, # label_var: reader_train.streams.labels #} C.logging.log_number_of_parameters(z) ; print() # perform model training batch_index = 0 plot_data = {'batchindex':[], 'loss':[], 'error':[]} for epoch in range(max_epochs): sample_count = 0 X_, Y_ = shuffle(x_train,y_train) for i in range(n_batches): start = i * minibatch_size end = start + minibatch_size trainer.train_minibatch({input_var:X_[start:end],label_var:Y_[start:end]}) sample_count += minibatch_size # count samples processed so far # For visualization... plot_data['batchindex'].append(batch_index) plot_data['loss'].append(trainer.previous_minibatch_loss_average) plot_data['error'].append(trainer.previous_minibatch_evaluation_average) batch_index += 1 trainer.summarize_training_progress() # # Evaluation action # epoch_size = 10000 minibatch_size = 16 n_batches = len(x_test) // minibatch_size # process minibatches and evaluate the model metric_numer = 0 metric_denom = 0 sample_count = 0 minibatch_index = 0 sample_count = 0 X_, Y_ = shuffle(x_test,y_test) for i in range(n_batches): start = i * minibatch_size end = start + minibatch_size metric_numer += trainer.test_minibatch({input_var:X_[start:end],label_var:Y_[start:end]})* minibatch_size #metric_denom += current_minibatch ミニバッチサイズ metric_denom += minibatch_size # Keep track of the number of samples processed so far. sample_count += minibatch_size # count samples processed so far minibatch_index += 1 print("") print("Final Results: Minibatch[1-{}]: errs = {:0.1f}% * {}".format(minibatch_index+1, (metric_numer*100.0)/metric_denom, metric_denom)) print("") # Visualize training result: window_width = 32 loss_cumsum = np.cumsum(np.insert(plot_data['loss'], 0, 0)) error_cumsum = np.cumsum(np.insert(plot_data['error'], 0, 0)) # Moving average. plot_data['batchindex'] = np.insert(plot_data['batchindex'], 0, 0)[window_width:] plot_data['avg_loss'] = (loss_cumsum[window_width:] - loss_cumsum[:-window_width]) / window_width plot_data['avg_error'] = (error_cumsum[window_width:] - error_cumsum[:-window_width]) / window_width plt.figure(1) plt.subplot(211) plt.plot(plot_data["batchindex"], plot_data["avg_loss"], 'b--') plt.xlabel('Minibatch number') plt.ylabel('Loss') plt.title('Minibatch run vs. Training loss ') plt.show() plt.subplot(212) plt.plot(plot_data["batchindex"], plot_data["avg_error"], 'r--') plt.xlabel('Minibatch number') plt.ylabel('Label Prediction Error') plt.title('Minibatch run vs. Label Prediction Error ') plt.show() return C.softmax(z) #In[13]: print('In[13]:') pred = train_and_evaluate(X_train, Y_train, X_test, Y_test, max_epochs=8, model_func=create_basic_model) print('save pred_basic_model_bn -> temp/Output/CNTK_201B_CIFAR-10_pred_8.cmf') pred.save('temp/Output/CNTK_201B_CIFAR-10_pred_8.cmf')

序に、評価用のプログラムのコードも書きました。

''' CNTK_201B_CIFAR-10_ImageHandsOn_eval_nishi2.py cntk-Tutorials-2-5/CNTK_201B_CIFAR-10_ImageHandsOn_nishi.py 用 の検証プログラムです。 Created on 2018/04/22 @author: nishi ''' #from __future__ import print_function import numpy as np import cntk as C import os from PIL import Image from cntk.device import try_set_default_device, gpu from cntk import load_model, placeholder, Constant from cntk import Trainer from cntk.logging.graph import find_by_name, get_node_outputs from cntk.io import MinibatchSource, ImageDeserializer, StreamDefs, StreamDef import cntk.io.transforms as xforms from cntk.layers import Dense from cntk.learners import momentum_sgd, learning_parameter_schedule, momentum_schedule from cntk.ops import combine, softmax from cntk.ops.functions import CloneMethod from cntk.losses import cross_entropy_with_softmax from cntk.metrics import classification_error from cntk.logging import log_number_of_parameters, ProgressPrinter import matplotlib.pyplot as plt base_folder = os.path.dirname(os.path.abspath(__file__)) tl_model_file = os.path.join('temp','Output','CNTK_201B_CIFAR-10_pred_8.cmf') print('tl_model_file',tl_model_file) #_data_folder = os.path.join('C:\\tmp\\cntk-examples-2-4','data', 'CIFAR-10') #_train_map_file = os.path.join(_data_folder, "train_map.txt") #_test_map_file = os.path.join(_data_folder, "test_map.txt") from scipy.misc import toimage from keras.datasets import cifar10 from sklearn.utils import shuffle ''' keras.dataset.cifar10 ライブラリを使って CIFAR-10データセットをロード cifar10.load_data() は、一度データをダウンロードすると、 どこかにキャッシュが残っていて、2回目からは、ダウンロードが省略される様です。 ''' (X_train, Y_train), (X_test, Y_test) = cifar10.load_data() print(X_train.shape, Y_train.shape) print(X_test.shape, Y_test.shape) #print('Y_train[:2]') #print(Y_train[:2]) Y_train = np.eye(10)[Y_train.astype(int)] # 1-of-k表現に変換 Y_test = np.eye(10)[Y_test.astype(int)] # 1-of-k表現に変換 X_train=X_train.astype(np.float32) X_test_cnt=X_test.astype(np.float32) Y_train=Y_train.astype(np.float32) Y_test=Y_test.astype(np.float32) # X_train.shape (50000,32,32,3) -> (50000,3,32,32) に変換が必要です。 X_train= np.rollaxis(X_train,3,1) X_test_cnt= np.rollaxis(X_test_cnt,3,1) print(X_train.shape, Y_train.shape) print(X_test.shape, Y_test.shape) #print('Y_train[:2]') #print(Y_train[:2]) def eval_single_image(pred_op, image_data): #print('image_data.shape=',image_data.shape) arguments = {pred_op.arguments[0]:[image_data]} output = pred_op.eval(arguments) #print('output=',output) print('output=',np.max(output)) sm = softmax(output[0]) return sm.eval() def eval(): label_lookup = ["airplane", "automobile", "bird", "cat", "deer", "dog", "frog", "horse", "ship", "truck"] i=0 for i in range(100): img_cnt = X_test_cnt[i] true_label=np.argmax(Y_test[i]) probs = eval_single_image(trained_model,img_cnt) if probs[0]=='None': print("Eval not possible: ") continue print("true_label=",true_label,',',label_lookup[true_label]) predicted_label = np.argmax(probs) print("predicted_label=",predicted_label,',',label_lookup[predicted_label]) img = X_test[i] plt.imshow(img) plt.show() i+=1 if os.path.exists(tl_model_file): print("Loading existing model from %s" % tl_model_file) trained_model = load_model(tl_model_file) if False: print("print trained_model node_outputs") node_outputs = get_node_outputs(trained_model) for out in node_outputs: print("{0} {1}".format(out.name, out.shape)) else: print("model not found") print("exit()") exit() eval()

これで、ディスクアクセス100% に悩まされる事無く、モデルを学習させて、分類のテストが出来るようになりました。

分類の結果を画像を見ながら感じたのですが、犬と猫など分類が難しい様な気がします。
分類の結果が、間違いやすい物だった場合、例えば、犬か猫が出てきた場合は、 cifar-2(犬と猫) と言う、２段目の学習モデルを用意してそちらを、再度通して分類するのは、どうか?
その場合、1段目の学習モデルは、粗い画像を使い、２段目の学習モデルには、精細な画像を使って精度を上げて行くのはどうか?
でも、２段目も畳み込みをしたら、同じなので、２段目はそれ以外の方法では?
おんちゃんの考えは、以上ぞね。