#coding:utf-8

import os

import chainer
from chainer import optimizers
import chainer.functions as F
import chainer.links as L
import chainer.serializers as S

import numpy as np

class model2(chainer.Chain):

    def __init__(self):
	
        n_input = 9248
        n_units = 2000
		
        super(model2, self).__init__(
            conv1 =  F.Convolution2D(3, 16, 5, pad=2),
            conv2 =  F.Convolution2D(16, 32, 5, pad=2),
            l1 =  F.Linear(n_inputs, n_units),           
            l2 =  F.Linear(n_units, 14) ,
        )

    def clear(self):
        self.loss = None
        self.accuracy = None

    def forward(self, X_data, y_data, train=True):
        self.clear() 
        X_data = chainer.Variable(np.asarray(X_data), volatile=not train)
        y_data = chainer.Variable(np.asarray(y_data), volatile=not train)
        h = F.max_pooling_2d(F.relu(self.conv1(X_data)), ksize = 5, stride = 2, pad =2)
        h = F.max_pooling_2d(F.relu(self.conv2(h)), ksize = 5, stride = 2, pad =2)
        h = F.dropout(F.relu(self.l1(h)), train=train)
        y = self.l2(h)
        return F.softmax_cross_entropy(y, y_data), F.accuracy(y, y_data)