mlp_clf_mnist_test

import os
os.environ['CUDA_VISIBLE_DEVICES'] = "0"
from mlp_clf import MLPClassifier
import numpy as np
import tensorflow as tf
from sklearn.datasets import load_svmlight_file
from sklearn.utils import shuffle
#from scipy.sparse import csr_matrix
def mean(numbers): #计算平均值
s = 0.0
for num in numbers:
s = s + num
return s/len(numbers)
def dev(numbers, mean): #计算方差
sdev = 0.0
for num in numbers:
sdev = sdev + (num - mean)**2
return pow(sdev / (len(numbers)-1), 0.5)

if __name__ == '__main__':
mze_array=[]
mae_array=[]
mse_array=[]

trainfile = "test_housing.0"
testfile = "train_housing.0"

X_train0, y_train = load_svmlight_file(trainfile)
X_test0 , y_test = load_svmlight_file(testfile)
X_train = X_train0.todense()
X_test = X_test0.todense()
attribute = np.shape(X_train)[-1]#tf.shape(X_train)
n_class = tf.reduce_max(y_train)
print("n_class:", n_class)

X_train, y_train = shuffle(X_train, y_train, random_state=0)

y_train = y_train - 1
y_test = y_test - 1

clf = MLPClassifier(attribute, 5, [100]*3)
log = clf.fit(X_train, y_train, n_epoch=1000, keep_prob=0.8, val_data=(X_test, y_test))
Y_pred = clf.predict(X_test)
mze = 1- (Y_pred == y_test).mean()
mae = np.abs(Y_pred.ravel()-y_test.ravel()).astype(float).mean()
mse = np.power(Y_pred.ravel()-y_test.ravel(),2).astype(float).mean()

print("mze: %.4f" % mze)
print("mae:", mae)
print("mse:", mse)

Rudolf Wille. Restructuring lattice theory: An approach based on hierarchies of concepts. In Sébastien Ferré andSebastian Rudolph, editors,Formal Concept Analysis, pages 314–339, Berlin, Heidelberg, 2009. Springer BerlinHeidelberg.

R. Wille. Restructuring lattice theory: An approach based on hierarchies of concepts. In
I. Rival, editor, Ordered Sets, pages 445–470. Reidel, Dordrecht-Boston, 1982.