Memory error in lomb_scargle.py

[vberten]

Hi there,

I've been using Cesium successfully since several months in Python 3.5, but I upgraded my system to 3.6 recently (and Cesium 0.9.9), and I can't make Cesium working anymore.

I got the following MemroyError :
[...]

/opt/anaconda3/envs/python36/lib/python3.6/site-packages/cesium/features/lomb_scargle.py in fit_lomb_scargle(time, signal, error, f0, df, numf, nharm, psdmin, detrend_order, freq_zoom, tone_control, lambda0, lambda0_range)
    171     hat_hat = np.zeros((npar,npar),dtype='float64')
    172     soln = np.zeros(npar,dtype='float64')
--> 173     psd = np.zeros(numf,dtype='float64')
    174 
    175     # Detrend the data and create the orthogonal detrending basis

MemoryError: 

By adding a "print(numf)" in lomb_scargle.py , I get "648648000000000000", which might explain the error (but I don't explain why I get now this number).

I'm using a "multi-channel" dataset, and some of my time series contain just zero's

I have this error even if I call featurize_time_series with features_to_use=["mean"]. I find a little bit weird that fit_lomb_scargle is called anyway.

Any clue ? Need more info ?

Thanks!

[stefanv]

That number looks too big, for sure!

Can you try to produce a small, self-contained code snippet that triggers the error?

Also, which platform are you on (Mac, Linux, Win)?

[vberten]

Hi Stephan, Here is a minimal example, with data in a pickle file, and outputs in a log file. I’m running this on a Linux platform, but tried also on Windows (Anaconda) with the same result. Cesium 0.9.9, dask 1.0.0, gatspy 0.3, numpy 1.15.4 [CesiumTest.zip](https://github.com/cesium-ml/cesium/files/2782365/CesiumTest.zip)

[stefanv]

Thanks, @vberten, we will investigate. /cc @acrellin

[stefanv]

I cannot reproduce; @vberten are you still experiencing this problem?

[stefanv]

(Perhaps also try v0.10.1, just released, in which I have fixed a few memory allocation issues.)