search

blue-yonder / tsfresh

Automatic extraction of relevant features from time series:
http://tsfresh.readthedocs.io
MIT License
8.4k stars 1.21k forks source link

About the freezing of tsfresh even at built-in data set #400

Closed aws-haoyuli closed 6 years ago

aws-haoyuli commented 6 years ago
  1. Your operating system Windows 10 & Ubuntu 16.04 LTS
  2. The version of tsfresh that you are using 0.11
  3. The data on which the problem occurred (please do not upload 1000s of time series but try to boil the problem down to a small group or even a singular one) I used the robot_execution_failures(built-in data set) and one my own data set. The form is as follows: index id time EMG BVP GSR RESP 0 1 0 5.02 24.83 5.42 46.41
  4. A minimal code snippet which reproduces the problem/bug 
    extract_settings = ComprehensiveFCParameters()
X = extract_features(df, column_id='id', column_sort='time', default_fc_parameters=extract_settings, impute_function=impute)
X_filtered = extract_relevant_features(df, y,column_id='id', column_sort='time', default_fc_parameters=extract_settings)

I run this code with Python 3.6 on Windows 10 OS and it is always stuck whichever data set I use. After a long time like more than 10 hours, it displays 'MemoryErr'. I run the code with Python 2.7 on Ubuntu 16.04 LTS and it can finish correctly on robot_execution_failures data set but it is stuck at 80% at my own data set.

  1. Any reported errors or traceback It is just stuck.

For questions, you can also use our gitter chatroom

MaxBenChrist commented 6 years ago

how big is your dataset?

can you try minimal settings on your dataset?

aws-haoyuli commented 6 years ago

My dataset contains 80 instances and every instance contains 2001 rows so it contains about 160000 rows. On Linux I tried thousands of rows and I can get correct result. But for 20, 40 and 80 instances I can't get any results. The RAM may be too small? Because I only allocate 2 GB RAM for it. On Windows I never get any results on any dataset of any size.

MaxBenChrist commented 6 years ago

None of the developers has access to a windows machine, so we can only give rudimentary support for tsfresh on windows.

Do you have any logs or error messages that pop up?

As said, please try to run your feature extraction with MinimalFCParameters or EfficientFCParameters, not the ComprehensiveFCSettings. The comprehensiveFCSettings contain some features that may not converge or take a long time to calculate for certain types of time series.

nicholasg97 commented 5 years ago

I just wanted to say that I was having the same issue on a windows machine within an Anaconda environment, and what solved the issue for me was uninstalling tsfresh using pip and installing with conda install -c conda-forge tsfresh

gammaci commented 5 years ago

I just wanted to say that I was having the same issue on a windows machine within an Anaconda environment, and what solved the issue for me was uninstalling tsfresh using pip and installing with conda install -c conda-forge tsfresh

Thanks, this worked for me too, side note no need to uninstall with pip, just overwritten previous installation.