Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Ilyas Moutawwakil
    @IlyasMoutawwakil
    on prophet's github it says here that fbprophet is kinda ghosted:
    Version 1.0 (2021.03.28)
    Python package name changed from fbprophet to prophet
    Fixed R Windows build issues to get latest version back on CRAN
    Improvements in serialization, holidays, and R timezone handling
    Plotting improvements
    Chaim Yosef Glancz
    @chaimglancz
    The major dependency that prophet has is pystan and I have windows and pystan has problems by me with windows so I use only the old version with the name fbprophet so I'm happy that sktime also uses the old version
    Franz Király
    @fkiraly
    yes, indeed - we've tried upgrading but it broke so many things for us developers and users, we decided it might be such a major time and resource investment to upgrade it that we would focus on other things. Already the original version and its dependencies was really a hassle, see also this issue alan-turing-institute/sktime#666
    We've also been thinking of moving the Bayesian dependencies like pystan out of sktime core since they continue causing problems for users, but again that might be a major undertaking.
    Either way, contributions are very much appreciated - but be aware that this is a challenging and potentially frustrating task...
    here's one try by Martin alan-turing-institute/sktime#841, by all I know he decided to take a long holiday after that.
    Siddhi Shroff
    @Sidshroff
    CIF
    Hi, everyone! I am trying to implement CanonicalIntervalForest from sktime
    I was able to implement Catch22classfier with the same dataset, but when I implement CIF I get an error! Will somebody be able to help me with it?
    Siddhi Shroff
    @Sidshroff
    This is the main error:
    ~/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/sktime/classification/base.py in fit(self, X, y)
    108 X, y = check_X_y(X, y, coerce_to_numpy=coerce_to_numpy)
    109
    --> 110 self._fit(X, y)
    111
    112 # this should happen last
    Matthew Middlehurst
    @MatthewMiddlehurst
    @Sidshroff Hi. Unfortunetly, that just tells us that it fails in fit somewhere. All the tests pass for CIF.
    Does your data have unequal length series? Our classifiers arent really set up to handle that currently.
    Siddhi Shroff
    @Sidshroff
    @MatthewMiddlehurst - Thanks for the response Matthew. The data is equal length series, I handled that during the data per-processing part. Is there anything I can look at to figure out whats causing the error?
    Matthew Middlehurst
    @MatthewMiddlehurst
    @Sidshroff Yeah, would be good to find out so we can fix it if required. If the data is equal length with no missing values/NaNs it should be fine.
    Matthew Middlehurst
    @MatthewMiddlehurst
    Is there an exception/error given other than that? A bit difficult if its only telling you where the issue is.
    I could try running it if i had the data, but understandable if you cant share
    Siddhi Shroff
    @Sidshroff
    Let me check, and see if I can share more details on the error. Sorry, it would be difficult to share the data, i really do appreciate the help!
    Siddhi Shroff
    @Sidshroff
    image.png
    Sorry, it doesn't look very readable, understandable if you would be unable to help with this error message
    Matthew Middlehurst
    @MatthewMiddlehurst
    No thats great, thought i had already fixed that. Will try push somthing in a couple of days.
    Siddhi Shroff
    @Sidshroff
    Would you be able to tell me whats causing the error? It would good for me to know
    Matthew Middlehurst
    @MatthewMiddlehurst
    i can send you somthing now if you can edit the code, if not try a different random seed.
    Siddhi Shroff
    @Sidshroff
    Sure thing! I can definitely try that, in the meanwhile I will try changing the random seed as well
    Matthew Middlehurst
    @MatthewMiddlehurst
    its trying to randomly select a length, but there is only one possible length to select.
    so its trying to generate a random number between 0 and 0 (which it cant do)
    Siddhi Shroff
    @Sidshroff
    Ah, that makes sense! And this is due to the fact that all my time series (I have 20 variables) of the same length? Thanks so much!
    Matthew Middlehurst
    @MatthewMiddlehurst
    you can include these lines after line 277, which should work
                    if len_range - self._min_interval > 0
                    else self._min_interval
    no, as your series is short it has a low amount of possible lengths
    Siddhi Shroff
    @Sidshroff
    Got it! Sure, let me try this!
    Matthew Middlehurst
    @MatthewMiddlehurst
    i fixed one part of it before, but neglected the other. my bad!
    Will include that in a PR later.
    Siddhi Shroff
    @Sidshroff
    Thanks for the quick turnaround time on this! All good :-). I am really excited to implement CIF and check ML performance lift for my dataset. Also, really enjoyed the paper on it!
    Siddhi Shroff
    @Sidshroff
    @MatthewMiddlehurst - Thank you for the tip! I was able to run the CIF on my dataset. Just wanted to check if it's normal for predict_proba on CIF to take a long time?
    For me, its almost been 45 mins!
    Hancel PV
    @hancelpv
    Hi All, I wanted to use the ExpSmoothing model with the lower and upper bounds for alpha, beta, gamma, phi.
    I can't see this functionality being implemented in sktime, whereas the underlying statsmodels estimator supports this
    Has anyone faced similar issue ? Please suggest solutions if any. Worst case I can re use the statsmodels model itself.
    Hancel PV
    @hancelpv
    Sorry about this, I found the solution, I should have used AutoETS instead of ExpSmoothing class.
    Ilyas Moutawwakil
    @IlyasMoutawwakil
    Hello everyone, I know this is not the subject of the chat but can anyone here give me an advice about spatio-temporal forecasting (event data), some python packages or algorithmes I could use for this task?
    Bhaskar Dhariyal
    @haskarb
    image.png
    Is this the best way to use 1NN-DTW in sktime? Is there other faster implementation?
    Matthew Middlehurst
    @MatthewMiddlehurst
    That's how you would run it, yes. We are currently in the process of updating the distance measures.
    1 reply