search

apple / turicreate

Turi Create simplifies the development of custom machine learning models.
BSD 3-Clause "New" or "Revised" License
11.19k stars 1.14k forks source link

Time Series manipulation functionality in Turi Create #378

Open jen0liu opened 6 years ago

jen0liu commented 6 years ago

They are in the in the original graphlab package. It seems like quite a few features were peeled off from the recent Turi release?

mxlei01 commented 6 years ago

+1

I'm working on 20TBs of time series data, and sframes does not have resampling methods, which requires me to iterate over chunks of sframes and converting them to pandas dataframe, then perform resampling, and then append it back to sframes.

JaDuminy commented 6 years ago

Do we know when and if the time series datatype feature request will be added into a future build?

rterbush commented 6 years ago

Struggling with missing TimeSeries method as well.

srikris commented 6 years ago

Using this issue to track TimeSeries only.

srikris commented 6 years ago

@jen0liu @rterbush @JaDuminy Can you elaborate on your use cases a bit so that we can build some more clarity around this?

foreversunyao commented 4 years ago

Any progress updates for this ? do we have any plan for when will this new feature be released ?

syoutsey commented 4 years ago

Hi @foreversunyao, we don't have any immediate details regarding support for this feature. Could you give us more detail about your use case and how you'd like to use this feature? Thanks!

foreversunyao commented 4 years ago

@syoutsey the features like https://turi.com/products/create/docs/generated/graphlab.TimeSeries.html , resample and others, right now I have to use pandas to do this part

akrishnamo commented 4 years ago

Hi. Any update on the gl.TimeSeries data structure coming back? Is it being addressed by feature enhancements to SFrame or SArray instead?

roganchris commented 4 years ago

I just discovered Turicreate doesn't support timeseries structures that Graphlab Create did. I developed my baseline code on a Windows machine using Graphlab Create and am in the process of porting it over to a Linux server with Turicreate. During the baseline development I initially used pandas dataframes for timeseries downsampling due to the multi-indexing capabilities but converted it over to graphlab timeseries structures due to the size of the datasets, exceeding 400m records, 35GB sframes.

Most of my analysis is timeseries driven with the need to downsample and interpolate data from second granularity to daily, weekly, monthly and quarterly samplings. We use this data for historical trend analysis and projecting forward. It very disappointing and honestly unexpected that this core data structure is no longer supported.