hpac / linnea Goto Github PK

I get the following error when I call graph.write_output with experiment_code=True.

  File "/usr/lib/python3.6/site-packages/pkg_resources/__init__.py", line 1465, in _get
    return self.loader.get_data(path)
OSError: [Errno 0] Error: 'linnea/code_generation/experiments/templates/Cpp/CMakeLists.txt'

Running the example script from examples/run_linnea.py fails with the following exception:

Traceback (most recent call last):
  File "/Users/user/Code/linnea/examples/run_linnea.py", line 4, in <module>
    import linnea.config
  File "/Users/user/Code/linnea/linnea/config.py", line 247, in <module>
    load_config()
  File "/Users/user/Code/linnea/linnea/config.py", line 237, in load_config
    set_output_path(value)
  File "/Users/user/Code/linnea/linnea/config.py", line 157, in set_output_path
    raise DirectoryDoesNotExist(output_path)
linnea.config.DirectoryDoesNotExist: /Users/user/path/to/output

Note that examples/config.json sets the output path to "~/path/to/output" which might not necessarily exist. Not also however, that the output path option would be overridden in the code to be ".", but the example crashes before it comes to that.
From reading the README.md it sounds like since options from a config.json can be overridden via linnea.config.set_X that a non existing directory should not crash the example.

While using the script 'examples/run_linnea.py' for a random equation over and over again, is it possible to set a seed so that Linnea generates the same set of variants in the same order every time?

I use this script to generate Julia codes. I get incorrect code (non-deterministic) for some Application examples.

Some errors:

Application 05

Application 16 (operand sizes from "operand_generator.jl" and the requirements in generated code does not match

Also it would be better if the variable id numbers in the generated code are reset every time graph.write_output is called

Why not default no_duplicates=True? or Is there any reason to generate duplicate code?

Change the logic of MatrixGenerators. Strip them down of all checks/responsibilities and provide different/dedicated constructors for every matrix combination style. This will enable us to fine-tune parameters such as eigenvalues etc. based on the operand properties (Triangular, SPD etc.).

Reference here

Dear All,
I would be interested in using the Linnea frontend to parse a specific expression and emit intermediate representation (compiler IR). Can I use the class Equation to do that? (

Hi there,

First of all, great tool!

Out of curiosity, given that Linnea produces Julia code eventually, why isn't it also implemented in Julia (rather than Python)? Any particular considerations?

Best,
Carsten

For the random expression the in script below, I always find only 1 algorithm when I set k_best arg to True in graph.write_output. But when I set it to False, I always find 5 algorithms.

    import random
    import linnea.config

    linnea.config.set_output_code_path(".")
    linnea.config.init()

    from linnea.derivation.graph.derivation import DerivationGraph
    from linnea.examples.random_expressions import generate_equation

    random.seed(0)
    rand_exprs = [generate_equation(random.randint(4, 7)) for _ in range(100)]
    equations = rand_exprs[33]

    graph = DerivationGraph(equations)
    graph.derivation(time_limit=60,
                     merging=True,
                     dead_ends=True,
                     pruning_factor=1.5)

    graph.write_output(code=True,
                       derivation=False,
                       output_name="tmp",
                       k_best=True,
                       experiment_code=False,
                       algorithms_limit=100,
                       graph=False)

I'm getting dimension mismatch for this example while executing the runner script.

        A = Matrix("A", (m, n))

        B = Matrix("B", (n, k))

        C = Matrix("C", (k, l))

        D = Matrix("D", (k, l))

        Y = Matrix("Y", (m, l))

        # Y = AB(C+ D)
        
        self.eqns = Equations(Equal(Y,Plus(Times(A,B,C),Times(A,B,D))))

Shouldn't this be M*N?

• Export all individual timings from function benchmarks (Julia, Matlab, Cpp)
• Create python scripts that process those timings and compute confidence intervals.