Python examples fixes

[breznak]

Fixes #712

[breznak]

OT: this file is a good place to test and replace the python likelihood with TM.anomaly. Currently the Likelihood from TM (c++) really gives bad results. #665

[t0n8]

    plt.plot(np.arange(len(inputs)), inputs, 'red',
             np.arange(len(inputs)), anomaly, 'blue',)

Line 232/233 throws the following error at this end:
ValueError: x and y must have same first dimension, but have shapes (4385,) and (4391,)

This seems to be a side effect from line 194
inputs = inputs[6:] #crop the first max prediction-steps inputs (as those don't have inferences)

So maybe line 232/233 should be changed to:

    plt.plot(np.arange(len(inputs)), inputs, 'red',
             np.arange(len(inputs)), anomaly[6:], 'blue',)

Would that make sense?

[dkeeney]

Looks good to me.

[breznak]

This seems to be a side effect from line 194
inputs = inputs[6:] #crop the first max prediction-steps inputs (as those don't have inferences)

@t0n8 Are you reviewing up to date sources? a) that still shouldn't be a problem since both arrays have been cropped, b) the change is now gone with bb3808b

[breznak]

Thank you for review, David
@t0n8 let us know if everything works, it should now

[t0n8]

Hi @breznak,

Tried again and do get a different error:

ERR:  CHECK FAILED: "dimensions_ != 0" Classifier: must call `learn` before `infer`. [D:\a\htm.core\htm.core\src\htm\algorithms\SDRClassifier.cpp line 50]
Traceback (most recent call last):
  File "hotgym.py", line 240, in <module>
    main()
  File "hotgym.py", line 161, in main
    pdf = predictor.infer( tm.getActiveCells() )
RuntimeError: CHECK FAILED: "dimensions_ != 0" Classifier: must call `learn` before `infer`.

Now I am wondering, whether this has to do with the fact that I am using the updated hotgym.py code with the current binary release.

I'll give it another try, once a new binary has been released.

Sorry for any confusion this might have caused.

Sorry for any confusion.

[breznak]

whether this has to do with the fact that I am using the updated hotgym.py code with the current binary release.

yes, that would be the case. For now you'll need to build from source to get the fix (see Readme on how to do so), or wait till a new release.