Discussions

Ask a Question

Access to embeddings

Does the product offer access to the transformer's embeddings? My use cases are different from forecasting or anomaly detection (the only capabilities mentioned in the docs). Am I stuck?

Timeout Issue with fine_tune_steps >= 370 on TimeGPT

I'm experiencing timeouts when setting fine_tune_steps >= 370 during model training with the following configuration: - **Dataset Size**: 17k rows - **Model**: timegpt-1-long-horizon - **Key Parameters**: - fine_tune_steps=400 - clean_ex_first=False - id_col="id" - add_history=False - df=df_exog1 - h=48 - target_col='--------' - time_col="DateTime_x" - freq="min" Despite having sufficient resources in Google Colab, the process times out consistently when fine_tune_steps is set to 370 or higher. **What could be causing these timeouts, and is there a recommended solution? Are there any subscription plans or higher-performance APIs available that can handle more extensive fine-tuning without these issues?**

How to fetch API key?

I am trying to fetch an API key but receiving a white blank page in response while clicking on 'Get Api Key' tab. Can anyone please help me with this?

Anomaly detection

while plotting anomaly, how can we know the exact date of the plotted anomaly.

Is this product still active? Is it being shut down/discontinued?

It seems like it is nearly shut down.

Is this product still active? Is it being shut down/discontinued?

It seems like it is nearly shut down.

Is there a complete example including precise API KEY description?

Which API key? What? Nobody knows? Where are the time columns? No time columns? Is it invariant to column name permutations? Is it invariant to column permutations? How to identify the target predictor? What is the task form? P(Y'|X, y, X') is not supported?

NIXTLA plots

Following the quick start guide, these plots do not appear when calling nixtla_client.plot() Could you please add more information on IDEs preferred, environments, and additional dependencies? Running in a conda environment there are dependency problems with numpy and matplotlib. Notebooks will not work, from this approach, and the VS Code IDE will not display plots.

endogenous Variables

how to feed endogenous variables to help in forecasting ?, forecast function call can only see options for feeding exogenous variables.

does nixtla predict well for day-wise prediction

does nixtla predict well for day-wise prediction, because im getting a huge difference in the actual and predicted values .