How to use the joblib.func_inspect.filter_args function in joblib

output_dir: string
                directory where to write metadata.

            duration: float
                time taken by hashing input arguments, calling the wrapped
                function and persisting its output.

            args, kwargs: list and dict
                input arguments for wrapped function

            this_duration_limit: float
                Max execution time for this function before issuing a warning.
        """
        start_time = time.time()
        argument_dict = filter_args(self.func, self.ignore,
                                    args, kwargs)

        input_repr = dict((k, repr(v)) for k, v in argument_dict.items())
        # This can fail due to race-conditions with multiple
        # concurrent joblibs removing the file or the directory
        metadata = {"duration": duration, "input_args": input_repr}

        func_id, args_id = self._get_output_identifiers(*args, **kwargs)
        self.store_backend.store_metadata([func_id, args_id], metadata)

        this_duration = time.time() - start_time
        if this_duration > this_duration_limit:
            # This persistence should be fast. It will not be if repr() takes
            # time and its output is large, because json.dump will have to
            # write a large file. This should not be an issue with numpy arrays
            # for which repr() always output a short representation, but can

def get_output_dir(self, *args, **kwargs):
        """ Returns the directory in which are persisted the results
            of the function corresponding to the given arguments.

            The results can be loaded using the .load_output method.
        """
        coerce_mmap = (self.mmap_mode is not None)
        argdict = filter_args(self.func, self.ignore,
                             args, kwargs)
        for k,v in self.hashfun.iteritems():
            if k in argdict:
                argdict[k] = v(argdict[k])
        #import ipdb; ipdb.set_trace()
        argument_hash = hash(argdict,
                             coerce_mmap=coerce_mmap)
        output_dir = os.path.join(self._get_func_dir(self.func),
                                  argument_hash)
        return output_dir, argument_hash

def get_output_key(self, args, kwargs):
        """Return the key that the output should be cached with, given
        arguments, keyword arguments, and a list of arguments to ignore.
        """
        # Get a dictionary mapping argument names to argument values where
        # ignored arguments are omitted.
        filtered_args = joblib.func_inspect.filter_args(
            self.func, self.ignore, args, kwargs
        )
        # Get a sorted tuple of the filtered argument.
        filtered_args = tuple(sorted(filtered_args.values()))
        # Use native hash when hashing arguments.
        return db.generate_key(filtered_args)

def _persist_input(self, output_dir, *args, **kwargs):
        """ Save a small summary of the call using json format in the
            output directory.
        """
        argument_dict = filter_args(self.func, self.ignore,
                                    args, kwargs)

        input_repr = dict((k, repr(v)) for k, v in argument_dict.iteritems())
        if json is not None:
            # This can fail do to race-conditions with multiple
            # concurrent joblibs removing the file or the directory
            try:
                mkdirp(output_dir)
                json.dump(
                    input_repr,
                    file(os.path.join(output_dir, 'input_args.json'), 'w'),
                    )
            except:
                pass
        return input_repr