'Series' object has no attribute 'ds'. Did you mean: 'xs'

[Yag-ger]

Student & first GitHub Post ever. I hope this is the right issue in the right place. Feel free to move this post.

I am using Nixtlats TimeGPT in Python and I already created a token on the website. The token also works, the validation function gives back true. Upon calling the function detect_anomalies on a simple timeseries the following error is shown:

AttributeError: 'Series' object has no attribute 'ds'. Did you mean: 'xs'?

I tried to find a solution in the docs but couldnt find any. I also tried to rename my date column to ds. Still no Luck.

Logs:

outliers_timegpt = timegpt.detect_anomalies(timeseries, time_col='ds', target_col='prognosebasis', freq='W') ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\ygerkman\AppData\Local\miniconda3\Lib\site-packages\nixtlats\timegpt.py", line 1190, in detect_anomalies return self._detect_anomalies( ^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\ygerkman\AppData\Local\miniconda3\Lib\site-packages\nixtlats\timegpt.py", line 625, in wrapper return func(self, *args, kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\ygerkman\AppData\Local\miniconda3\Lib\site-packages\nixtlats\timegpt.py", line 643, in wrapper return func(self, kwargs, num_partitions=1) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\ygerkman\AppData\Local\miniconda3\Lib\site-packages\nixtlats\timegpt.py", line 818, in _detect_anomalies anomalies_df = timegpt_model.detect_anomalies(df=df) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\ygerkman\AppData\Local\miniconda3\Lib\site-packages\nixtlats\timegpt.py", line 527, in detectanomalies df, = self.transform_inputs(df=df, X_df=None) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\ygerkman\AppData\Local\miniconda3\Lib\site-packages\nixtlats\timegpt.py", line 188, in transform_inputs if df.dtypes.ds != "object": ^^^^^^^^^^^^ File "C:\Users\ygerkman\AppData\Local\miniconda3\Lib\site-packages\pandas\core\generic.py", line 5902, in getattr return object.getattribute(self, name) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ AttributeError: 'Series' object has no attribute 'ds'. Did you mean: 'xs'?

[grace001214]

你好，我也遇到了这个问题，想问下您解决这个问题了么，如何解决的呢

[Yibei990826]

Hi @Yag-ger and @grace001214. I was not able to replicate this problem, but I think it might be a column naming issue. Could you please rename the timestamp column and let me know if the problem is still persistent?

[jmoralez]

I think there's a bug here https://github.com/Nixtla/nixtla/blob/7ceb2532c0911086dd27285ef281a9d10d63cfd3/nixtlats/timegpt.py#L193-L206

If the name of the index is 'hello' and the TimeGPT object is initialized like TimeGPT(time_col='ds') (the default) it'll try to rename 'ds': 'ds' which won't do anything since that column won't exist, it should instead rename 'hello': 'ds'

[Yibei990826]

I think there's a bug here

https://github.com/Nixtla/nixtla/blob/7ceb2532c0911086dd27285ef281a9d10d63cfd3/nixtlats/timegpt.py#L193-L206

If the name of the index is 'hello' and the TimeGPT object is initialized like TimeGPT(time_col='ds') (the default) it'll try to rename 'ds': 'ds' which won't do anything since that column won't exist, it should instead rename 'hello': 'ds'

I agree with you, we will see this issue anytime the date col is passed as index, (even though it might be named 'ds'). For instance, I was able to replicate the issue with the following data:

I think it might be helpful to add something like:

elif hasattr(df.index, "freq") and  hasattr(df.index, "name"):
     if df.index.name = self.time_col:
        df=df.reset_index()