def get_model(train_file,
model_file,
ndim=784,
nout=10,
val_fraction=0.3,
islearning=True):
'''
'''
args = parse_args(train_file, model_file, islearning, val_fraction)
if args.network == 'mlp':
data_shape = (ndim, )
net = get_mlp_spice(nout)
elif args.network == 'cnn':
data_shape = (15, 30) #sentence_size, num_embed
net = get_cnn_spice(num_label=nout, batch_size=args.batch_size)
# train
model = models.fit(args, net, get_iterator_spice(data_shape))
model = load_obj(model_file)
return model, args.network
def fit(store=None, data=None, **kwargs):
return models.fit(Configuration(**kwargs), DataContext(store, data))
def fit(store=None, data=None, **kwargs):
return models.fit(Configuration(**kwargs), DataContext(store, data))
if not os.path.exists(path + 'output\\'):
os.mkdir(path + 'output\\')
os.chdir(path + "output\\")
result.to_csv(filename, index=None)
zip = zipfile.ZipFile(
filename.split('.')[0] + '.zip', "w", zipfile.ZIP_DEFLATED)
zip.write(filename)
zip.close()
if __name__ == "__main__":
'''提供了2018年1月1日至2018年12月31日的标的第一期的还款数据作为训练集，
需要选手预测2019年2月1日至2019年3月31日成交标的第一期的还款情况'''
#时间序列，回归，或者分类(排除逾期的用户，剩下用户直接统计还款日作为结果) 二分类或多分类（逾期，逾期前1天、2天、3天还款）
args = parse_command_params()
# analysis_data(args['path'])
train, test = gen_data(args)
print(train.shape, test.shape)
cols = list(
set(train.columns) - set([
'user_id', 'listing_id', 'auditing_date', 'due_date', 'insertdate',
'label'
]))
model = models.fit(train, args, cols)
result = gen_result(model, test, cols)
save_zip(result, args)
print('done !')
# For songs sampling
"TEMPERATURE": 1,
"TAKE_MAX_PROBABLE": False,
"LIMIT_LEN": 300
}
print(config)
# model = VanillaRNN(config["VOCAB_SIZE"], config["HIDDEN"], config["VOCAB_SIZE"]).to(get_device())
model = LSTMSimple(config["VOCAB_SIZE"], config["HIDDEN"],
config["VOCAB_SIZE"]).to(get_device())
criterion = CrossEntropyLoss()
# Fit Model
fit(model, train_encoded, val_encoded, config)
# Report NLL for validation and test
nll_val = negative_log_likelihood(model, val_encoded, criterion, config)
nll_test = negative_log_likelihood(model, test_encoded, criterion, config)
print("NLL Validation: {}".format(nll_val))
print("NLL Test: {}".format(nll_test))
# Save error plot to file
save_loss_graph(model)
# Save model to file
print("Saving model...")
now = datetime.now().strftime('%Y-%m-%d-%H-%M')
torch.save(model.state_dict(), "model" + now + ".pth")
print("Saved!")
def get_reward(actions, dataset):
reward = models.fit(actions, dataset)
return reward
def get_reward(self, actions):
reward = models.fit(actions, self.dataset)
return reward