def LSTMCell(xt, h, c):
f = edf.Sigmoid(edf.Add(edf.VDot(edf.ConCat(xt, h), Wf), bf))
i = edf.Sigmoid(edf.Add(edf.VDot(edf.ConCat(xt, h), Wi), bi))
o = edf.Sigmoid(edf.Add(edf.VDot(edf.ConCat(xt, h), Wo), bo))
c_hat = edf.Tanh(edf.Add(edf.VDot(edf.ConCat(xt, h), Wc), bc))
c_next = edf.Add(edf.Mul(f, c), edf.Mul(i, c_hat))
h_next = edf.Mul(o, edf.Tanh(c_next))
return h_next, c_next
def LSTMCell(xt, h, c, layer):
f = edf.Sigmoid(edf.Add(edf.VDot(edf.ConCat(xt, h), Wf[layer]), bf[layer]))
i = edf.Sigmoid(edf.Add(edf.VDot(edf.ConCat(xt, h), Wi[layer]), bi[layer]))
o = edf.Sigmoid(edf.Add(edf.VDot(edf.ConCat(xt, h), Wo[layer]), bo[layer]))
c_hat = edf.Tanh(edf.Add(edf.VDot(edf.ConCat(xt, h), Wc[layer]), bc[layer]))
c_next = edf.Add(edf.Mul(f, c), edf.Mul(i, c_hat))
h_next = edf.Mul(o, edf.Tanh(c_next))
return h_next, c_next
def LSTMCell(x, h, c):
concat = edf.ConCat(h, x)
# Forget Gate
f_gate = edf.Sigmoid(edf.Add(edf.VDot(concat, Wf), bf))
# Input Gate
i_gate = edf.Sigmoid(edf.Add(edf.VDot(concat, Wi), bi))
# Temp Vars
c_temp = edf.Tanh(edf.Add(edf.VDot(concat, Wc), bc))
o_temp = edf.Sigmoid(edf.Add(edf.VDot(concat, Wo), bo))
# Output
c_next = edf.Add(edf.Mul(f_gate, c), edf.Mul(i_gate, c_temp))
h_next = edf.Mul(o_temp, edf.Tanh(c_next))
return h_next, c_next
def BuildModel():
edf.components = []
B, T = inp.value.shape
score = []
loss = None
# Init h_0 with one-hot
vocab_init = np.ones([B])
vocab_init = edf.Value(vocab_init)
h = edf.Embed(vocab_init, C2V)
# Init C_0 to be zero
c = edf.Value(np.zeros([B, hidden_dim]))
for t in range(T):
x_t = edf.Value(inp.value[:, t])
x_t = edf.Embed(x_t, C2V)
h, c = LSTMCell(x_t, h, c)
# Score and loss
pred = edf.SoftMax(edf.VDot(h, V))
if t != T - 1:
score.append(pred)
x_t1 = edf.Value(inp.value[:, t + 1])
else:
x_t1 = edf.Value(np.zeros(B))
loss_t = edf.LogLoss(edf.Aref(pred, x_t1))
if loss is None:
loss = loss_t
else:
loss = edf.Add(loss, loss_t)
loss = edf.Mul(edf.Mean(loss), edf.Value(np.float64(1) / T))
return loss, score
# evaluation bucket bucket = 100 ########################## Simple Convolution Nerual Network Model for Cifar 10 ################################## ################################################################################################################## # please implement your main cnn model here, as described by the homework, you can mimic the previous code f1 = edf.Param(edf.xavier((3, 3, prev_channel, 32))) b1 = edf.Param(np.zeros(32)) f3 = edf.Param(edf.xavier((3, 3, 32, 64))) b3 = edf.Param(np.zeros(64)) f5 = edf.Param(edf.xavier((1, 1, 64, 10))) b5 = edf.Param(np.zeros(10)) layer1 = edf.RELU(edf.Add(Conv(f1, inp, 1, 1), b1)) layer2 = MaxPool(layer1, 4) layer3 = edf.RELU(edf.Add(Conv(f3, layer2), b3)) layer4 = AvePool(layer3, 6) layer5 = edf.RELU(edf.Add(Conv(f5, layer4), b5)) pred = edf.Reshape(layer5, (bucket, 10)) # the standard classification layer, which you don't need to modify pred = edf.SoftMax(pred) loss = edf.Mean(edf.LogLoss(edf.Aref(pred, lab))) acc = edf.Accuracy(pred, lab) ################################################################################################################## def eval_train():