CNTK_103C_MNIST_MultiLayerPerceptronの改造

''' CNTK_103C_MNIST_MultiLayerPerceptron_nishi.py オリジナルの CNTK_103C_MNIST_MultiLayerPerceptron をもとに データ入力部分を、 sklearn を使ってみます。 Created on 2018/05/30 @author: nishi ''' import matplotlib.image as mpimg import matplotlib.pyplot as plt import numpy as np import sys import os import cntk as C import cntk.ops import cntk.tests.test_utils from gc import callbacks cntk.tests.test_utils.set_device_from_pytest_env() # (only needed for our build system) C.cntk_py.set_fixed_random_seed(1) # fix a random seed for CNTK components #%matplotlib inline from sklearn import datasets #from sklearn.cross_validation import train_test_split from sklearn.model_selection import train_test_split from ctypes.test.test_pickling import name #from random import shuffle from sklearn.utils import shuffle ''' MNIST データ準備 ''' mnist = datasets.fetch_mldata('MNIST original',data_home='/tmp/MNIST_mnist_prg1') print(mnist.keys()) n = len(mnist.data) print('n=',n) N = 40000 # MNISTの一部のデータで実験。件数はここで増やして indices = np.random.permutation(range(n))[:N] # ランダムにN枚を選択 X = mnist.data[indices] y = mnist.target[indices] Y = np.eye(10)[y.astype(int)] # 1-of-k表現に変換 ''' 学習用=0.8 と テスト用=0.2 に分ける ''' #X_train,X_test,Y_train,Y_test = train_test_split(X,Y,train_size=0.8) X_train,X_test,Y_train,Y_test = train_test_split(X,Y,test_size=0.2) print(len(X_train)) print(len(X_test)) ''' データ、ラベルのデータ型を、CNTK2.5 の形式 np.float32 に変換 ''' X_train=X_train.astype(np.float32) #X_train= X_train/255.0 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=',X_train.shape) print('X_train.dtype=',X_train.dtype) print('Y_train.shape=',Y_train.shape) print('Y_train.dtype=',Y_train.dtype) #In[5]: print('In[5]:') # Define the data dimensions input_dim = 784 num_output_classes = 10 #In[8]: print('In[8]:') num_hidden_layers = 2 hidden_layers_dim = 400 #In[9]: print('In[9]:') input = C.input_variable(input_dim) # 入力 Feature label = C.input_variable(num_output_classes) # 正解ラベル ''' Multi-layer Perceptron setup ''' #In[10]: print('In[10]:') def create_model(features): with C.layers.default_options(init = C.layers.glorot_uniform(), activation = cntk.ops.relu): h = features for _ in range(num_hidden_layers): h = C.layers.Dense(hidden_layers_dim)(h) r = C.layers.Dense(num_output_classes, activation = None)(h) return r z = create_model(input) ''' 誤差関数: クロスエントロピー誤差関数 ''' #In[12]: print('In[12]:') loss = C.cross_entropy_with_softmax(z, label) ''' Evaluation ''' #In[13]: print('In[13]:') label_error = C.classification_error(z, label) ''' Configure training ''' #In[14]: print('In[14]:') # 学習率は、活性化関数が、 relu の場合は、小さめにします。ここは、トライ・アンド・エラーで決めます。 learning_rate = 0.001 #lr_schedule = C.learning_parameter_schedule(learning_rate) lr_schedule = C.learning_rate_schedule(learning_rate, C.UnitType.minibatch) learner = C.sgd(z.parameters, lr_schedule) ''' トレーニング方法における注意点 sklearn による MNISTデータを使っています。 メモリー上の変数として、取得されているので、これをミニバッチデータとして trainer.train_minibatch(),trainer.test_minibatch()の入力に与える必要がありますが、 その方法は、CNTK_101_LogisticRegression.ipynb を参考にします。 ''' trainer = C.Trainer(z,(loss,label_error),[learner]) #progress_writer = C.logging.ProgressPrinter(0) ''' CNTK_101_LogisticRegression.ipynb の 評価ツール ''' # Define a utility function to compute the moving average. # A more efficient implementation is possible with np.cumsum() function def moving_average(a, w=10): if len(a) < w: return a[:] return [val if idx < w else sum(a[(idx-w):idx])/w for idx, val in enumerate(a)] # Define a utility that prints the training progress def print_training_progress(trainer, mb, frequency, verbose=1): training_loss, eval_error = "NA", "NA" if mb % frequency == 0: training_loss = trainer.previous_minibatch_loss_average eval_error = trainer.previous_minibatch_evaluation_average if verbose: print ("Minibatch: {0}, Loss: {1:.4f}, Error: {2:.2f}".format(mb, training_loss, eval_error)) return mb, training_loss, eval_error ''' モデル学習 ''' #epochs = 1000 epochs = 6 batch_size = 50 # ミニバッチサイズ n_batches = len(X_train) // batch_size print('n_batches=',n_batches) training_progress_output_freq= 100 for epoch in range(epochs): print('epoch=',epoch) X_, Y_ = shuffle(X_train,Y_train) for i in range(n_batches): start = i * batch_size end = start + batch_size trainer.train_minibatch({input:X_[start:end],label:Y_[start:end]}) batchsize, loss, error = print_training_progress(trainer, i, training_progress_output_freq, verbose=1) evals = trainer.test_minibatch({input:X_test,label:Y_test}) print('evals=',evals) ''' 実際の判定 ''' out = C.softmax(z) ans = out.eval({input:X_test[:30]}) #print('ans') print("Ans :",[np.argmax(ansx) for ansx in ans]) print("Label:",[np.argmax(labelx) for labelx in Y_test[:30]])