Rubi support #1
Replies: 0 comments 27 replies
-
|
Thanks for undertaking this! For what it's worth, when I run And I see: two times. |
Beta Was this translation helpful? Give feedback.
-
|
Based on a first quick look, this seems like a performance issue with handling patterns with default values in Python functions ALso, if we let it run for a long time with just |
Beta Was this translation helpful? Give feedback.
-
|
@rocky Unfortunately, I think the problems are more basic: In[1]:= f[(a_. + b_. x_Symbol)^n_.] = 1
Out[1]= 1
# Ok
In[2]:= f[2 x]
Out[2]= 1
# Wrong!
In[3]:= f[1 + 2 x]
Out[3]= f[1 + 2 x]
# Wrong!
In[4]:= f[(1 + 2 x)^1]
Out[4]= f[1 + 2 x]
# Ok!
In[5]:= f[(0 + 2 x)^1]
Out[5]= 1
In[6]:= |
Beta Was this translation helpful? Give feedback.
-
|
@aravindh-krishnamoorthy It took me a while to understand the WMA-specific pattern-matching gobbledygook above. Although you did a good job isolating the problem somewhat, for me, a WMA novice, to begin to be able to fix the problems, it would help to break this down more into simpler, or even better the the simplest cases. Let me describe my understanding of the situation here in the hope that in the future we can break down problems into the kinds of things that someone who doesn't understand the WMA language. The more things are narrowed, the easier it is to address problems. For context, I gather this was motivated by the Here, what we have is a pattern-matching problem. These are more easily worked on using the So I gather that "+" (is this pattern or ?) is failing when a is 1 but not 0. It looks like the Can I go further, and remove the space (" ") in Changing to But at any rate, this is enough for me to start investigating the pattern-matching bugs we have. I will write up a bug report in mathics-core soon. |
Beta Was this translation helpful? Give feedback.
-
|
@rocky, notice that |
Beta Was this translation helpful? Give feedback.
-
|
Ok - thanks. In @aravindh-krishnamoorthy thanks for correcting that |
Beta Was this translation helpful? Give feedback.
-
|
@axkr You got this working for Symja, right? Do you have any advice on the process of getting a large package like this working? Or specific advice for debugging this package? Thanks. |
Beta Was this translation helpful? Give feedback.
-
We've taken a complete other way, which is described here: @halirutan helped in getting the Rubi rules in this special format: If you go this way, IMO you should try |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the information. Very interesting. It will take a while to look over and fully understand. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you, axkr and rocky for the inputs! My first attempt would be to understand the above issues and try and get the official Rubi running without any changes (since Rubi runs as-is in Mathematica). If it doesn't work at all, then perhaps we'd have to "port" the rules and/or change some of the conditions in "IntegrationUtilityFunctions". |
Beta Was this translation helpful? Give feedback.
-
|
One special thing I remember was, that I have to add the |
Beta Was this translation helpful? Give feedback.
-
|
@aravindh-krishnamoorthy we should definitely use (adapted for Python) the Integration test suite that Symja has developed. See https://github.com/axkr/symja_android_library/blob/master/symja_android_library/matheclipse-io/src/test/java/org/matheclipse/core/rubi/AlgebraicFunctions.java @axkr One thing that confuses me about these tests is that they call |
Beta Was this translation helpful? Give feedback.
-
Comparing only the strings for tests will not work for complicated expressions. I try to use my (not very perfect version) of |
Beta Was this translation helpful? Give feedback.
-
|
I see, thank you for the inputs, @axkr. I will keep this in mind. Right now, I'm debugging very basic things like matching default patterns, where I'm running into trouble. But, once this is set, I expect that we can begin integrating the algebraic tests. |
Beta Was this translation helpful? Give feedback.
-
Interesting. I guess I didn't mean "modified for Python" so much as not in Java format. I assumed this was hand generated since it is checked into git. But what format were the tests originally? If WMA, that is okay. What I guess I'd like happen is that unless this is already in WMA, it be put into some language-agnostic format like YAML. (YAML is slightly better than JSON because it supports adding comments.) And then this can be in a common repository were Mathics3 could read the YAML into Python and run it that way. The common YAML could have a field to indicate not to run for a particular project like Mathics3 or Symja. |
Beta Was this translation helpful? Give feedback.
-
The tests are somehow extracted from the testsuite: |
Beta Was this translation helpful? Give feedback.
-
|
@aravindh-krishnamoorthy Since the test suite from Rubi is provided in WMA format, let's try to use that. |
Beta Was this translation helpful? Give feedback.
-
|
I can give some insight into what exactly we did when testing Symja. You can test three things when you want to compare Mathics behavior to how Rubi works in Mathematica:
You can find here how this looks in Symja, although I seem to remember we only used 1-step integration tests in this case. When we ran into issues with an integration problem and dissected each Rubi expression, we created a unit test like this. I have two accompanying and undocumented Mathematica notebooks that I used back then, and I'm happy to share them and provide some help. One notebook is for creating unit tests that compare expressions or rule numbers on integration problems and exports a whole integration test suite. The other notebook shows how to trace Rubi calls and create unit tests for each evaluation happening during an integration. Who do you think would be the best person to move this forward? |
Beta Was this translation helpful? Give feedback.
-
As happens with open-source projects, it is often unclear who is the "best" person. Everybody has strengths and weaknesses, and everyone is volunteering as well. So I can only speak for myself: I can't use a Mathematica notebook. (The project is not far enough along that we can read these in Mathics3). And as is evident in this discussion, my knowledge of how WMA works is a bit limited. A suggestion then is just to use this discussion; everyone who is interested, then will see. |
Beta Was this translation helpful? Give feedback.
-
Let's say |
Beta Was this translation helpful? Give feedback.
-
|
@aravindh-krishnamoorthy I have created https://github.com/Mathics3/Mathics3-Rubi and have invited you as a maintainer. I'd be grateful if you would work there instead of under the fork/branch in mathics core. Instead you can checkout this git repo as a git submodule. Thanks. |
Beta Was this translation helpful? Give feedback.
-
This is a general thread about current Rubi support.
Current status
Note
A very small subset of Rubi is usable. See below for instructions.
1 Algebraic functions1.1 Binomial products1.1.1 Linear1.1.1.1 (a+b x)^m1.1.1.2 (a+b x)^m (c+d x)^n(.........There are nine more sections besides
1 Algebraic functions. So, a long way to go!Usage instructions
Beta Was this translation helpful? Give feedback.
All reactions