pycnn xor实例

本篇以经典的xor为例，解释使用pycnn的完整流程。

from pycnn import *hidden_size = 8iterations = 500m = Model()sgd = SimpleSGDTrainer(m)m.add_parameters('W', (hidden_size, 2))m.add_parameters('b', hidden_size)m.add_parameters('V', (1, hidden_size))m.add_parameters('a', 1)renew_cg() # new computation graph. not strictly needed here, but good practice.W = parameter(m['W'])b = parameter(m['b'])V = parameter(m['V'])a = parameter(m['a'])x = vecInput(2)y = scalarInput(0)h = tanh((W*x)+b)y_pred = logistic((V*h)+a)loss = binary_log_loss(y_pred, y)x_examples = [[0, 0], [0, 1], [1, 0], [1, 1]]y_examples = [0, 1, 1, 0]for i in xrange(iterations):    mloss = 0.0    for j in xrange(len(x_examples)):        x.set(x_examples[j])        # 为模型参数赋值        y.set(y_examples[j])        mloss += loss.scalar_value()    # 该步会执行正向传播forward        loss.backward()                 # 执行反向传播，计算参数的梯度        sgd.update(1.0)                 # 更新模型参数，Here 1.0 is the scaling factor that allows us to control the size of the update.    # sgd.update_epoch()    mloss /= 4    if i % (iterations/10) == 0 or i == (iterations-1):        print 'iter %d, loss: %f' % (i, mloss)for i in xrange(len(x_examples)):    x.set(x_examples[i])    print '[%d, %d]: %f' % (x_examples[i][0], x_examples[i][1], y_pred.scalar_value())

参考资料

1. pycnn-api
2. pycnn-examples-xor