Open ozanbarism opened 2 months ago
Hi @ozanbarism; the scaling might be the issue. A quick fix is to turn off scaling. Can you try that? Please also follow issue https://github.com/time-series-foundation-models/lag-llama/issues/85 which is related.
Hi, I am trying to test Lagllama on a prediction task. the dataset i feed is what i use for context. then using that context, i want it to make predictions. however, the first step of the predictions is very off from the last value of the context window. Why might this be the case?
This is my code. class LagLlamaModel: def init(self, checkpoint_path="/Users/ozanbaris/Documents/GitHub/TS-foundation-model/lag_src/lag-llama.ckpt"): self.checkpoint_path = checkpoint_path self.device = torch.device('cpu') self.scaler = MinMaxScaler()