Not working on Max 8 (Mac OSX Yosemite - 10.10.5)

[gustavosol]

Dear Jamie Bullock
First of all, thank you for this incredible work. I used this library on my PhD on Pd and was amazing.
I'm trying to use this library on Max 8 (2.2 GHz Intel Core i7 - Mac OSX Yosemite - 10.10.5) with no success.
Those main objects doesn't open [svm] and [mlp], even when I try with the help file.
The address to find the external seems to be correct because works with others.
I attached an image.

Could you help me, please?

[sandcobainer]

Same problem here, I upgraded from Max 7 to 8 to workaround the probabilities feature. I think it is the 64 bit/ 32 bit problem here. Did you try starting max as 32 bit?

[gustavosol]

Hum, I've just tried the 32 bit (Max 7.3.5) and no results. The problem still there.
Maybe I'm doing something wrong in the moment to saving the external or files, like max path or something regarding library...

[jamiebullock]

Hi, I can't reproduce this here. It works in Max 8 64-bit fine for me.

Can you try this build: https://www.dropbox.com/s/gf1v8k3frbv24t5/ml-0.19.0-alpha-OSX64-2.zip?dl=0

And let me know if there is still a problem or not? If the problem persists, please post any error messages from the Max console. Thanks!

[jamiebullock]

Hi @gustavosol, did the updated build resolve this?

[gustavosol]

Dear Jamie Look at this video: when I try to train, it crashes. Thank you! MaxCrash.mov <https://drive.google.com/file/d/1HkyfxLMHGeO_-pH8yfehiR3e6a5k7-Yu/view?usp=drive_web>

…

On Thu, Feb 28, 2019 at 4:15 PM Jamie Bullock ***@***.***> wrote: Hi @gustavosol <https://github.com/gustavosol>, did the updated build resolve this? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#151 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL885zKseS_JZab2dql7nmam1k0VCU_Zks5vSCrqgaJpZM4ZTeKn> .

-- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

[NiccoloGranieri]

After having spent a while trying to reproduce the crash, I think I have some useful information both to prevent ml.lib from crashing and for a development fix.

@gustavosol
It looks like the message you are feeding into ml.mlp is formatted like this:

(add 60. 0.5 60. 1.8 0. 0. accx accy accz)

The formatting of the message tells ml.mlp that you want to add a class with the id of 60. of 8 features (0.5 60. 1.8 0. 0. accx accy accz). Because the different classes are probably expected to be with an incremental id, ml.mlp crashes when it doesn't find all the classes with id lower than 60 (from 1 to 59).

Because I don't exactly understand the importance of the values that precede the accxyz, I am going to propose two solutions. If the 60 right after add is meant to be the class identifier, change it to 1:

(add 1 0.5 60. 1.8 0. 0. accx accy accz)

otherwise, to leave the 9 number feature list intact add a 1 in between add and 60.:

(add 1 60. 0.5 60. 1.8 0. 0. accx accy accz)

@jamiebullock
It seems to me that ml.lib crashes when ml.mlp is fed a class with an id > 2 not preceded by all the classes with lower id but > 1. The crash seems to be happening only with ml.mlp.

To reproduce:
Add to ml.mlp the example class changing the id with any number > 3.
Press train.

Crash Report - ml.mlp

[gustavosol]

Thank you very much. I'll try your suggestion and send you a feedback in the next week. All the best. Gustavo

…

On Tue, Mar 5, 2019 at 2:05 PM Niccolò ***@***.***> wrote: After having spent a while trying to reproduce the crash, I think I have some useful information both to prevent ml.lib from crashing and for a development fix. @gustavosol <https://github.com/gustavosol> It looks like the message you are feeding into ml.mlp is formatted like this: (add 60. 0.5 60. 1.8 0. 0. accx accy accz) The formatting of the message tells ml.mlp that you want to add a class with the id of 60. of 8 features (0.5 60. 1.8 0. 0. accx accy accz). Because the different classes are probably expected to be with an incremental id, ml.mlp crashes when it doesn't find all the classes with id lower than 60 (from 1 to 59). Because I don't exactly understand the importance of the values that precede the accxyz, I am going to propose two solutions. If the 60 right after add is meant to be the class identifier, change it to 1: (add 1 0.5 60. 1.8 0. 0. accx accy accz) otherwise, to leave the 9 number feature list intact add a 1 in between add and 60.: (add 1 60. 0.5 60. 1.8 0. 0. accx accy accz) @jamiebullock <https://github.com/jamiebullock> It seems to me that ml.lib crashes when ml.mlp is fed a class with an id > 2 not preceded by all the classes with lower id but > 1. The crash seems to be happening only with ml.mlp. *To reproduce:* Add to ml.mlp the example class changing the id with any number > 3. Press train. Crash Report - ml.mlp <https://github.com/irllabs/ml-lib/files/2931952/Max_2019-03-05-170357_Niccolos-MacBook-Pro.crash.zip> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#151 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL8855nhRpTtAmKm4Z4IfMy1HJmjmOBIks5vTqPfgaJpZM4ZTeKn> .

-- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

[gustavosol]

Dear Jamie, I got this message about the length of the input expected and found this object lost in the middle of the [p sound] object (in MLP). Thank you in advance. Gustavo [image: Screen Shot 2019-03-04 at 00.29.10.png] [image: Screen Shot 2019-03-04 at 00.03.07.png]

…

On Wed, Mar 6, 2019 at 12:38 PM Gustavo Sol ***@***.***> wrote: Thank you very much. I'll try your suggestion and send you a feedback in the next week. All the best. Gustavo On Tue, Mar 5, 2019 at 2:05 PM Niccolò ***@***.***> wrote: > After having spent a while trying to reproduce the crash, I think I have > some useful information both to prevent ml.lib from crashing and for a > development fix. > > @gustavosol <https://github.com/gustavosol> > It looks like the message you are feeding into ml.mlp is formatted like > this: > > (add 60. 0.5 60. 1.8 0. 0. accx accy accz) > > The formatting of the message tells ml.mlp that you want to add a class > with the id of 60. of 8 features (0.5 60. 1.8 0. 0. accx accy accz). > Because the different classes are probably expected to be with an > incremental id, ml.mlp crashes when it doesn't find all the classes with id > lower than 60 (from 1 to 59). > > Because I don't exactly understand the importance of the values that > precede the accxyz, I am going to propose two solutions. If the 60 right > after add is meant to be the class identifier, change it to 1: > > (add 1 0.5 60. 1.8 0. 0. accx accy accz) > > otherwise, to leave the 9 number feature list intact add a 1 in between > add and 60.: > > (add 1 60. 0.5 60. 1.8 0. 0. accx accy accz) > > @jamiebullock <https://github.com/jamiebullock> > It seems to me that ml.lib crashes when ml.mlp is fed a class with an id > > 2 not preceded by all the classes with lower id but > 1. The crash seems > to be happening only with ml.mlp. > > *To reproduce:* > Add to ml.mlp the example class changing the id with any number > 3. > Press train. > > Crash Report - ml.mlp > <https://github.com/irllabs/ml-lib/files/2931952/Max_2019-03-05-170357_Niccolos-MacBook-Pro.crash.zip> > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#151 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AL8855nhRpTtAmKm4Z4IfMy1HJmjmOBIks5vTqPfgaJpZM4ZTeKn> > . > -- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

-- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

[NiccoloGranieri]

Hi @gustavosol, it seems like the images have not been linked correctly to your message. Would you mind re-attaching them?

Thanks.

[gustavosol]

I tried again. Let's see if works now. Thank you for your support. Best

…

On Thu, Mar 7, 2019 at 7:13 PM Niccolò ***@***.***> wrote: Hi @gustavosol <https://github.com/gustavosol>, it seems like the images have not been linked correctly to your message. Would you mind re-attaching them? Thanks. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#151 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL8857-xuBe9sZEB6Vy0F680vBLmCx86ks5vUY7zgaJpZM4ZTeKn> .

-- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

[NiccoloGranieri]

Hi @gustavosol, unfortunately I still don't see any image. It might be a GitHub issue. When you drag and drop an image on the text box here on GitHub can you make sure the resulting line of text looks something like this?

exclamation_mark[Name of file in square brackets](Link_of_file_in_parenthesis)

Also, before clicking on comment, you can click on the Preview tab in the top left corner, close to the Write tab, to make sure the image has been correctly attached and there is a visible preview of said image.

[jamiebullock]

Hi @gustavosol @NiccoloGranieri ...

I think Niccolo's workaround should work for this, but I have also found the source of the bug.

Basically, it is an upstream bug in GRT. I have reported it here if you want to look at the details, but in essence the class labels for GRT must be contiguous starting from 1.

So you can't have class labels of 2, 3, 6, 9, they must be 1, 2, 3, 4. You cannot have 1, 2, 4, it must be 1, 2, 3 etc.

I may attempt to fix this and send an upstream PR.

[NiccoloGranieri]

Hi @gustavosol, @jamiebullock fixed the bug and now your patch should work out of the box without implementing any changes.

Could you try this latest release: https://www.dropbox.com/s/s3n4vtwve48bdgj/ml.lib-0.20.2.zip?dl=0

and see if that solves your issue?

[gustavosol]

Dear Nicolò, Jamie and ml lib team This is a great work! It is working almost well. Although, I'm still receiving length error like the image here. Thank you and all the best Sol

…

On Sat, Mar 9, 2019 at 9:57 AM Niccolò ***@***.***> wrote: Hi @gustavosol <https://github.com/gustavosol>, @jamiebullock <https://github.com/jamiebullock> fixed the bug and now your patch should work out of the box without implementing any changes. Could you try this latest release: https://www.dropbox.com/s/s3n4vtwve48bdgj/ml.lib-0.20.2.zip?dl=0 and see if that solves your issue? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#151 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL885w__F6i8sFVAQ-rR7vBo8cwitsGcks5vU6-sgaJpZM4ZTeKn> .

-- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

[gustavosol]

I've solved his. But I'm still facing problems to write or read the file as you can see in this exemple attached for both SVM and MLP Thank you Sol

…

On Tue, Mar 19, 2019 at 6:10 PM Gustavo Sol ***@***.***> wrote: Dear Nicolò, Jamie and ml lib team This is a great work! It is working almost well. Although, I'm still receiving length error like the image here. Thank you and all the best Sol On Sat, Mar 9, 2019 at 9:57 AM Niccolò ***@***.***> wrote: > Hi @gustavosol <https://github.com/gustavosol>, @jamiebullock > <https://github.com/jamiebullock> fixed the bug and now your patch > should work out of the box without implementing any changes. > > Could you try this latest release: > https://www.dropbox.com/s/s3n4vtwve48bdgj/ml.lib-0.20.2.zip?dl=0 > > and see if that solves your issue? > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#151 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AL885w__F6i8sFVAQ-rR7vBo8cwitsGcks5vU6-sgaJpZM4ZTeKn> > . > -- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

-- http://www.facebook.com/gustavo.sol.77 https://www.gustavosol.com.br/ Gustavo Sol 011-9614-6692

[NiccoloGranieri]

Hi @gustavosol.

I still can't see the attachment you are referring to... how weird!

I just tested the write / read functions on Max 8.0.3 and everything seems to work fine. Could you please try the following?

1. Open the help file for ml.svm (or ml.mlp)
2. add the example class (add 1 0.2 0.7 0.3 0.1)
3. change the write message so that it's pointing to a location on your mac. A good example that should work for you could be:
   write /Users/ML
   This will write the data in a file called "ML.data" in the Users folder of your mac.
   3.1 Write the data by pressing the write message that you just modified.
4. change the read message so that it's point to the same location where you wrote the data. Using the same example:
   read /Users/ML.data
   This will read the data stored in the "ML.data" file you previously wrote.
   4.1 Read the data by pressing the read message that you just modified.
5. Check the Max Console for any errors

If following this example still gives you problems, could you please attach the problematic files to your next example? A link to a google drive file like you previously did would be perfect.

Thanks