def lstm():
maxlen = 50
X, Y, Z = gettuple()
print(len(X), len(Y), len(Z))
print('begin')
nbatch = 64
outdim = 3
args = {
'emb_size': 300,
'hidden_size': 128,
'num_layers': 2,
'dropout': 0.5
}
BATCH_SIZE = 1
model = SeqLSTMClassify(args, outdim)
for i in range(0, len(X), nbatch):
up = min(len(X), i + nbatch)
seq_tensor1, y1, seq_lengths1, perm_idx1 = embed_and_pad(
X[i:up], Z[i:up], maxlen, 'int')
seq_tensor2, y2, seq_lengths2, perm_idx2 = embed_and_pad(
Y[i:up], Z[i:up], maxlen, 'int')
loss = model.learn_once(
((seq_tensor1, seq_lengths1), (seq_tensor2, seq_lengths2)), y1)
print(loss.data.cpu().numpy())
def train(x, y, maxlen, EPOCH=1):
outdim = 2
args = {
'emb_size': 300,
'hidden_size': 128,
'num_layers': 2,
'dropout': 0.5
}
BATCH_SIZE = 1
model = SeqLSTMClassify(args, outdim)
for epoch in range(EPOCH):
idx = np.arange(len(x), dtype=int)
np.random.shuffle(idx)
for i in range(0, len(x), BATCH_SIZE):
print(i)
seq_tensor, y, seq_lengths, perm_idx = embed_and_pad(
x[idx[i:i + BATCH_SIZE]], y[idx[i:i + BATCH_SIZE]], maxlen)
loss = model.learn_once((seq_tensor, seq_lengths), y)
print(loss.data.cpu().numpy())
return model
def predict(df, model):
seq_tensor, y, seq_lengths, perm_idx = embed_and_pad(
df, np.arange(len(df.index)))
return model.predict((seq_tensor, seq_lengths), y, perm_idx)
def eval(x, y, maxlen, model):
seq_tensor, y, seq_lengths, perm_idx = embed_and_pad(x, y, maxlen)
loss = model.learn_once((seq_tensor, seq_lengths), y)
print(loss.data.cpu().numpy())
return loss.data.cpu().numpy()
def eval(df, model):
seq_tensor, y, seq_lengths, perm_idx = embed_and_pad(
df, np.arange(len(df.index)))
loss = model.learn_once((seq_tensor, seq_lengths), y)
print(loss.data.cpu().numpy())
return loss.data.cpu().numpy()