Armaghan Shakir

> The only change in RNN/LSTM adaptations is that this PR is using `1.7.0` instead of `1.6.4`, and nothing else as far as I am aware. Yes the tests run...

To reproduce `neuralforecast==1.7.0` ```py import pandas as pd from sktime.forecasting.neuralforecast import NeuralForecastRNN y = pd.Series(data=range(10), index=pd.RangeIndex(start=0, stop=10)) model = NeuralForecastRNN( freq="W", # works fine if integer is provided here max_steps=5,...

> can you please check how "auto" behaviour would work with `RangeIndex`? It raises an expected Value Error. ``` ValueError in NeuralForecastRNN, could not interpret freq, try passing freq in...

> I have the test `test_neural_forecast_with_auto_freq` failing too locally with the error is it on 1.7.0?

> Will that only not satisfy a pass CI? I think we do need to change the class. I don't think that only satisfies a CI pass because the deal...

> We need to check with @benHeid first whether he is ok with this renaming. yes sure, we'll wait > `pre-commit` did not pick up the name. Also I found...

@fkiraly Can you please make yourself clear as I think there is a contradiction in 2 statements by you https://github.com/sktime/sktime/issues/5932#issuecomment-1960436557 > But there are many similar places in the documentation...

@fkiraly Thanks for the clarification. I am working on a script that locates all the incorrectly used single-grave. Although there are some places where we need the single-grave, yet they...

If a model is available on huggingface with trained weights - is it still required to create an interface for that model in sktime? For example, `Autoformer` can be loaded...

If I am not wrong, basically, on a conceptual level, this would be the right way to do that? ```py # config = AutoConfig.from_pretrained(self.model_path) from transformers import AutoformerConfig config =...