def evaluate(self):
"""
评估单个行为的uAUC值
"""
if self.stage in ["online_train", "offline_train"]:
# 训练集，每个action一个文件
action = self.action
else:
# 测试集，所有action在同一个文件
action = "all"
file_name = "{stage}_{action}_{day}_concate_sample.csv".format(
stage=self.stage, action=action, day=STAGE_END_DAY[self.stage])
evaluate_dir = os.path.join(FLAGS.root_path, self.stage, file_name)
df = pd.read_csv(evaluate_dir)
userid_list = df['userid'].astype(str).tolist()
predicts = self.estimator.predict(
input_fn=lambda: self.input_fn_predict(df, self.stage, self.action
))
# print(file_name)
# print(predicts)
predicts_df = pd.DataFrame.from_dict(predicts)
# logits = predicts_df["logistic"].map(lambda x: x[0])
print(predicts_df.head())
logits = predicts_df["logits"].map(lambda x: x[0])
labels = df[self.action].values
uauc = uAUC(labels, logits, userid_list)
return df[["userid", "feedid"]], logits, uauc
def evaluate(self, df):
#测试集
test_x = df[self.select_frts].values
labels = df[self.action].values
userid_list = df['userid'].astype(str).tolist()
logits = self.model.predict(test_x)
uauc = uAUC(labels, logits, userid_list)
return df[["userid", "feedid"]], logits, uauc
def test(model, data_loader, device):
model.eval()
targets, predicts, user_ids = list(), list(), list()
with torch.no_grad():
for cat, nu, feed, target in tqdm.tqdm(data_loader,
smoothing=0,
mininterval=1.0):
cat, nu, feed, target = cat.to(device).long(), nu.to(
device), feed.to(device), target.to(device)
y = model(cat, nu.unsqueeze(2), feed)
user_ids.extend(cat[:, 0].tolist())
targets.extend(target.tolist())
predicts.extend(y.tolist())
return uAUC(targets, predicts, user_ids)
def evaluate(self, x, y, userid_list, batch_size=256):
"""
:param x: Numpy array of test data (if the model has a single input), or list of Numpy arrays (if the model has multiple inputs).
:param y: Numpy array of target (label) data (if the model has a single output), or list of Numpy arrays (if the model has multiple outputs).
:param batch_size: Integer or `None`. Number of samples per evaluation step. If unspecified, `batch_size` will default to 256.
:return: Dict contains metric names and metric values.
"""
pred_ans = self.predict(x, batch_size)
# y=y.reshape(y.shape[0],)
# pred_ans=pred_ans.reshape(pred_ans.shape[0],)
uauc = uAUC(y, pred_ans, userid_list)
print('uauc:', y.shape, pred_ans.shape, uauc)
eval_result = {}
eval_result['auc'] = uauc
return eval_result
