Hello! Asking about Inference.

[Zero-coder]

Thanks for innovating such a splendid work！I am enjoying watching your paper and code.
My question is:
After training and saving the model, how can I inference ?

[nooshinmaghsoodi]

Hi,
In one of the scripts in scripts folder (based on the task you want to do) set ckpt_path to address of the model you have saved and is_training to zero.

[tribeband]

I am also wondering the inference on the anomaly detection tast. it is shown that the train data is being calculated during test.
combined_energy = np.concatenate([train_energy, test_energy], axis=0)
threshold = np.percentile(
combined_energy, 100 - self.args.anomaly_ratio)
print("Threshold :", threshold)

with this implemention, when dealing with the new data, train data should be calculated first?
how about deployment and predict?

[wbzhang233]

Hi, In one of the scripts in scripts folder (based on the task you want to do) set ckpt_path to address of the model you have saved and is_training to zero.

Thanks for your generous reply! I wonder how can I use UniTS model for forecasting, but I cannot find any instruction about this issue. There is no reference for user.