search

pymc-devs / pytensor

PyTensor allows you to define, optimize, and efficiently evaluate mathematical expressions involving multi-dimensional arrays.
https://pytensor.readthedocs.io
Other
308 stars 93 forks source link

BUG: Cannot access `ndims_params` from `rng_fn()` classmethod #866

Closed vandalt closed 4 weeks ago

vandalt commented 4 weeks ago

Describe the issue:

I'm trying to fix a deprecation warnings in celerite2 following the recent update, namely the one about ndims_params and ndim_supp. I updated the custom RandomVariable with a signature attribute.

However, the rng_fn classmethod (link here) in celerite2 used ndim_params, which is no longer available via cls because it is now created when creating a new instance. I saw that some distributions in PyMC use the private method _parse_gufunc_signature along with _class_or_instancemethod, but I was not sure this was the best way forward for code outside PyMC 3.

I'm not sure this migration/update issue is technically a bug, but any advice on how to update this code following the recent update would be welcome!

Reproducable code example:

See link above

Error message:

FAILED python/test/pymc/test_pymc_distribution.py::test_celerite_normal_support_point[t0-0.0-None-expected0] - AttributeError: type object 'CeleriteNormalRV' has no attribute 'ndims_params'

PyTensor version information:

floatX ({'float16', 'float64', 'float32'}) Doc: Default floating-point precision for python casts. Note: float16 support is experimental, use at your own risk. Value: float64 warn_float64 ({'pdb', 'warn', 'ignore', 'raise'}) Doc: Do an action when a tensor variable with float64 dtype is created. Value: ignore pickle_test_value (>) Doc: Dump test values while pickling model. If True, test values will be dumped with model. Value: True cast_policy ({'custom', 'numpy+floatX'}) Doc: Rules for implicit type casting Value: custom deterministic ({'default', 'more'}) Doc: If `more`, sometimes we will select some implementation that are more deterministic, but slower. Also see the dnn.conv.algo* flags to cover more cases. Value: default device (cpu) Doc: Default device for computations. only cpu is supported for now Value: cpu force_device (>) Doc: Raise an error if we can't use the specified device Value: False conv__assert_shape (>) Doc: If True, AbstractConv* ops will verify that user-provided shapes match the runtime shapes (debugging option, may slow down compilation) Value: False print_global_stats (>) Doc: Print some global statistics (time spent) at the end Value: False assert_no_cpu_op ({'pdb', 'warn', 'ignore', 'raise'}) Doc: Raise an error/warning if there is a CPU op in the computational graph. Value: ignore unpickle_function (>) Doc: Replace unpickled PyTensor functions with None. This is useful to unpickle old graphs that pickled them when it shouldn't Value: True Doc: Default compilation mode Value: Mode cxx () Doc: The C++ compiler to use. Currently only g++ is supported, but supporting additional compilers should not be too difficult. If it is empty, no C++ code is compiled. Value: /usr/bin/g++ linker ({'c|py', 'c', 'py', 'cvm', 'cvm_nogc', 'vm_nogc', 'vm', 'c|py_nogc'}) Doc: Default linker used if the pytensor flags mode is Mode Value: cvm allow_gc (>) Doc: Do we default to delete intermediate results during PyTensor function calls? Doing so lowers the memory requirement, but asks that we reallocate memory at the next function call. This is implemented for the default linker, but may not work for all linkers. Value: True optimizer ({'None', 'o1', 'merge', 'o2', 'fast_compile', 'fast_run', 'o3', 'unsafe', 'o4'}) Doc: Default optimizer. If not None, will use this optimizer with the Mode Value: o4 optimizer_verbose (>) Doc: If True, we print all optimization being applied Value: False on_opt_error ({'ignore', 'pdb', 'warn', 'raise'}) Doc: What to do when an optimization crashes: warn and skip it, raise the exception, or fall into the pdb debugger. Value: warn nocleanup (>) Doc: Suppress the deletion of code files that did not compile cleanly Value: False on_unused_input ({'ignore', 'warn', 'raise'}) Doc: What to do if a variable in the 'inputs' list of pytensor.function() is not used in the graph. Value: raise gcc__cxxflags () Doc: Extra compiler flags for gcc Value: cmodule__warn_no_version (>) Doc: If True, will print a warning when compiling one or more Op with C code that can't be cached because there is no c_code_cache_version() function associated to at least one of those Ops. Value: False cmodule__remove_gxx_opt (>) Doc: If True, will remove the -O* parameter passed to g++.This is useful to debug in gdb modules compiled by PyTensor.The parameter -g is passed by default to g++ Value: False cmodule__compilation_warning (>) Doc: If True, will print compilation warnings. Value: False cmodule__preload_cache (>) Doc: If set to True, will preload the C module cache at import time Value: False cmodule__age_thresh_use () Doc: In seconds. The time after which PyTensor won't reuse a compile c module. Value: 2073600 cmodule__debug (>) Doc: If True, define a DEBUG macro (if not exists) for any compiled C code. Value: False compile__wait () Doc: Time to wait before retrying to acquire the compile lock. Value: 5 compile__timeout () Doc: In seconds, time that a process will wait before deciding to override an existing lock. An override only happens when the existing lock is held by the same owner *and* has not been 'refreshed' by this owner for more than this period. Refreshes are done every half timeout period for running processes. Value: 120 ctc__root () Doc: Directory which contains the root of Baidu CTC library. It is assumed that the compiled library is either inside the build, lib or lib64 subdirectory, and the header inside the include directory. Value: tensor__cmp_sloppy () Doc: Relax pytensor.tensor.math._allclose (0) not at all, (1) a bit, (2) more Value: 0 lib__amblibm (>) Doc: Use amd's amdlibm numerical library Value: False tensor__insert_inplace_optimizer_validate_nb () Doc: -1: auto, if graph have less then 500 nodes 1, else 10 Value: -1 traceback__limit () Doc: The number of stack to trace. -1 mean all. Value: 8 traceback__compile_limit () Doc: The number of stack to trace to keep during compilation. -1 mean all. If greater then 0, will also make us save PyTensor internal stack trace. Value: 0 warn__ignore_bug_before ({'0.8.1', '1.0.3', '0.8', '0.4.1', 'None', '0.7', '0.10', '0.5', '1.0', '1.0.1', 'all', '0.3', '1.0.2', '1.0.5', '0.8.2', '0.6', '0.4', '0.9', '1.0.4'}) Doc: If 'None', we warn about all PyTensor bugs found by default. If 'all', we don't warn about PyTensor bugs found by default. If a version, we print only the warnings relative to PyTensor bugs found after that version. Warning for specific bugs can be configured with specific [warn] flags. Value: 0.9 exception_verbosity ({'high', 'low'}) Doc: If 'low', the text of exceptions will generally refer to apply nodes with short names such as Elemwise{add_no_inplace}. If 'high', some exceptions will also refer to apply nodes with long descriptions like: A. Elemwise{add_no_inplace} B. log_likelihood_v_given_h C. log_likelihood_h Value: low print_test_value (>) Doc: If 'True', the __eval__ of an PyTensor variable will return its test_value when this is available. This has the practical consequence that, e.g., in debugging `my_var` will print the same as `my_var.tag.test_value` when a test value is defined. Value: False compute_test_value ({'warn', 'ignore', 'off', 'raise', 'pdb'}) Doc: If 'True', PyTensor will run each op at graph build time, using Constants, SharedVariables and the tag 'test_value' as inputs to the function. This helps the user track down problems in the graph before it gets optimized. Value: off compute_test_value_opt ({'warn', 'ignore', 'off', 'raise', 'pdb'}) Doc: For debugging PyTensor optimization only. Same as compute_test_value, but is used during PyTensor optimization Value: off check_input (>) Doc: Specify if types should check their input in their C code. It can be used to speed up compilation, reduce overhead (particularly for scalars) and reduce the number of generated C files. Value: True NanGuardMode__nan_is_error (>) Doc: Default value for nan_is_error Value: True NanGuardMode__inf_is_error (>) Doc: Default value for inf_is_error Value: True NanGuardMode__big_is_error (>) Doc: Default value for big_is_error Value: True NanGuardMode__action ({'pdb', 'warn', 'raise'}) Doc: What NanGuardMode does when it finds a problem Value: raise DebugMode__patience () Doc: Optimize graph this many times to detect inconsistency Value: 10 DebugMode__check_c (>) Doc: Run C implementations where possible Value: True DebugMode__check_py (>) Doc: Run Python implementations where possible Value: True DebugMode__check_finite (>) Doc: True -> complain about NaN/Inf results Value: True DebugMode__check_strides () Doc: Check that Python- and C-produced ndarrays have same strides. On difference: (0) - ignore, (1) warn, or (2) raise error Value: 0 DebugMode__warn_input_not_reused (>) Doc: Generate a warning when destroy_map or view_map says that an op works inplace, but the op did not reuse the input for its output. Value: True DebugMode__check_preallocated_output () Doc: Test thunks with pre-allocated memory as output storage. This is a list of strings separated by ":". Valid values are: "initial" (initial storage in storage map, happens with Scan),"previous" (previously-returned memory), "c_contiguous", "f_contiguous", "strided" (positive and negative strides), "wrong_size" (larger and smaller dimensions), and "ALL" (all of the above). Value: DebugMode__check_preallocated_output_ndim () Doc: When testing with "strided" preallocated output memory, test all combinations of strides over that number of (inner-most) dimensions. You may want to reduce that number to reduce memory or time usage, but it is advised to keep a minimum of 2. Value: 4 profiling__time_thunks (>) Doc: Time individual thunks when profiling Value: True profiling__n_apply () Doc: Number of Apply instances to print by default Value: 20 profiling__n_ops () Doc: Number of Ops to print by default Value: 20 profiling__output_line_width () Doc: Max line width for the profiling output Value: 512 profiling__min_memory_size () Doc: For the memory profile, do not print Apply nodes if the size of their outputs (in bytes) is lower than this threshold Value: 1024 profiling__min_peak_memory (>) Doc: The min peak memory usage of the order Value: False profiling__destination () Doc: File destination of the profiling output Value: stderr profiling__debugprint (>) Doc: Do a debugprint of the profiled functions Value: False profiling__ignore_first_call (>) Doc: Do we ignore the first call of an PyTensor function. Value: False on_shape_error ({'warn', 'raise'}) Doc: warn: print a warning and use the default value. raise: raise an error Value: warn openmp (>) Doc: Allow (or not) parallel computation on the CPU with OpenMP. This is the default value used when creating an Op that supports OpenMP parallelization. It is preferable to define it via the PyTensor configuration file ~/.pytensorrc or with the environment variable PYTENSOR_FLAGS. Parallelization is only done for some operations that implement it, and even for operations that implement parallelism, each operation is free to respect this flag or not. You can control the number of threads used with the environment variable OMP_NUM_THREADS. If it is set to 1, we disable openmp in PyTensor by default. Value: False openmp_elemwise_minsize () Doc: If OpenMP is enabled, this is the minimum size of vectors for which the openmp parallelization is enabled in element wise ops. Value: 200000 optimizer_excluding () Doc: When using the default mode, we will remove optimizer with these tags. Separate tags with ':'. Value: optimizer_including () Doc: When using the default mode, we will add optimizer with these tags. Separate tags with ':'. Value: optimizer_requiring () Doc: When using the default mode, we will require optimizer with these tags. Separate tags with ':'. Value: optdb__position_cutoff () Doc: Where to stop earlier during optimization. It represent the position of the optimizer where to stop. Value: inf optdb__max_use_ratio () Doc: A ratio that prevent infinite loop in EquilibriumGraphRewriter. Value: 8.0 cycle_detection ({'regular', 'fast'}) Doc: If cycle_detection is set to regular, most inplaces are allowed,but it is slower. If cycle_detection is set to faster, less inplacesare allowed, but it makes the compilation faster.The interaction of which one give the lower peak memory usage iscomplicated and not predictable, so if you are close to the peakmemory usage, triyng both could give you a small gain. Value: regular check_stack_trace ({'off', 'warn', 'log', 'raise'}) Doc: A flag for checking the stack trace during the optimization process. default (off): does not check the stack trace of any optimization log: inserts a dummy stack trace that identifies the optimizationthat inserted the variable that had an empty stack trace.warn: prints a warning if a stack trace is missing and also a dummystack trace is inserted that indicates which optimization insertedthe variable that had an empty stack trace.raise: raises an exception if a stack trace is missing Value: off metaopt__verbose () Doc: 0 for silent, 1 for only warnings, 2 for full output withtimings and selected implementation Value: 0 metaopt__optimizer_excluding () Doc: exclude optimizers with these tags. Separate tags with ':'. Value: metaopt__optimizer_including () Doc: include optimizers with these tags. Separate tags with ':'. Value: unittests__rseed () Doc: Seed to use for randomized unit tests. Special value 'random' means using a seed of None. Value: 666 warn__round (>) Doc: Warn when using `tensor.round` with the default mode. Round changed its default from `half_away_from_zero` to `half_to_even` to have the same default as NumPy. Value: False profile (>) Doc: If VM should collect profile information Value: False profile_optimizer (>) Doc: If VM should collect optimizer profile information Value: False profile_memory (>) Doc: If VM should collect memory profile information and print it Value: False Doc: Useful only for the VM Linkers. When lazy is None, auto detect if lazy evaluation is needed and use the appropriate version. If the C loop isn't being used and lazy is True, use the Stack VM; otherwise, use the Loop VM. Value: None numba__vectorize_target ({'cpu', 'parallel', 'cuda'}) Doc: Default target for numba.vectorize. Value: cpu numba__fastmath (>) Doc: If True, use Numba's fastmath mode. Value: True numba__cache (>) Doc: If True, use Numba's file based caching. Value: True compiledir_format () Doc: Format string for platform-dependent compiled module subdirectory (relative to base_compiledir). Available keys: device, gxx_version, hostname, numpy_version, platform, processor, pytensor_version, python_bitwidth, python_int_bitwidth, python_version, short_platform. Defaults to compiledir_%(short_platform)s-%(processor)s- %(python_version)s-%(python_bitwidth)s. Value: compiledir_%(short_platform)s-%(processor)s-%(python_version)s-%(python_bitwidth)s Doc: platform-independent root directory for compiled modules Value: /home/vandal/.pytensor Doc: platform-dependent cache directory for compiled modules Value: /home/vandal/.pytensor/compiledir_Linux-6.9-arch1-1-x86_64-with-glibc2.39--3.12.4-64 blas__ldflags () Doc: lib[s] to include for [Fortran] level-3 blas implementation Value: -L/usr/lib/ -lopenblas blas__check_openmp (>) Doc: Check for openmp library conflict. WARNING: Setting this to False leaves you open to wrong results in blas-related operations. Value: True scan__allow_gc (>) Doc: Allow/disallow gc inside of Scan (default: False) Value: False scan__allow_output_prealloc (>) Doc: Allow/disallow memory preallocation for outputs inside of scan (default: True) Value: True

Context for the issue:

No response

ricardoV94 commented 4 weeks ago

Hi @vandalt the ndims_params is available from the instantiated RandomVariable Op, but not passed to the classmethod rng_fn indeed. Would it be a solution to just hardcode the ndims_params in your call to broadcast_params?

Also there's some inefficiency in your rng_fn. If size is provided (size is not None), you shouldn't have to broadcast the params together, it's enough to have your call below where you broadcast each param to size (there you are also hardcoding implicitly the ndims_params when you do param.shape[-n:] btw).

Only when size is not provided do you need to broadcast the params, as size is implicitly the broadcasted batch shape of all the parameters.

ricardoV94 commented 4 weeks ago

Here is an example how we're using it internally, also hard-coded: https://github.com/pymc-devs/pytensor/blob/5d4b0c4b9a1e478dda48e912ee708a9e557e9343/pytensor/tensor/random/basic.py#L1797-L1801

vandalt commented 4 weeks ago

Thanks! Hardcoding works and I opened a pr in Celerite2. I'll open a separate issue regarding inefficiencies in rng_fn().