multipers 2.2.3__cp311-cp311-win_amd64.whl

multipers/slicer.pyx.tp

@@ -0,0 +1,914 @@
+{{py:
+
+"""
+Vine and non-vine slicers.
+both type have the same interface, defined the slicer.pyx file
+"""
+
+
+import pickle
+
+with open("build/tmp/_slicer_names.pkl", "rb") as f:
+  slicers=pickle.load(f)
+
+
+}}
+
+from multipers.simplex_tree_multi import SimplexTreeMulti, SimplexTreeMulti_type
+import multipers
+from typing import Optional,Literal
+import multipers.io as mio
+import multipers.grids as mpg
+import threading
+import os 
+from joblib import Parallel, delayed
+
+from multipers.slicer cimport *
+from multipers.filtrations cimport *
+from multipers.filtration_conversions cimport *
+## TODO: these two are not needed, remove that by updating rank code.
+from multipers.point_measure import sparsify, rank_decomposition_by_rectangles
+
+import numpy as np
+cimport cython
+from libcpp.string cimport string
+# python_value_type = np.float32
+from typing import Union
+from cython.operator cimport dereference
+
+## WARNING : This is repeated in the pxd file ...
+python_indices_type=np.int32
+python_tensor_dtype = np.int32
+
+global available_slicers
+available_slicers = tuple((
+{{for D in slicers}}
+  {{D['PYTHON_TYPE']}},
+{{endfor}}
+))
+
+global available_columns
+available_columns = set((
+{{for D in slicers}}
+  "{{D['COLUMN_TYPE']}}",
+{{endfor}}
+))
+
+global available_dtype
+available_dtype = set([
+{{for D in slicers}}
+  "{{D['PY_VALUE_TYPE']}}",
+{{endfor}}
+])
+
+
+global available_pers_backend
+available_pers_backend = set([
+{{for D in slicers}}
+  "{{D['PERS_BACKEND_TYPE']}}",
+{{endfor}}
+])
+
+
+
+global column_type
+_column_type = Literal[
+{{for D in slicers}}
+  "{{D['COLUMN_TYPE']}}",
+{{endfor}}
+]
+
+global _slicers_type
+Slicer_type = Union[
+{{for D in slicers}}
+  {{D['PYTHON_TYPE']}},
+{{endfor}}
+]
+
+global _valid_dtypes
+_valid_dtype = Union[
+{{for D in slicers}}
+  {{D['PY_VALUE_TYPE']}},
+{{endfor}}
+]
+
+
+global _valid_pers_backend
+_valid_pers_backend = Literal[
+{{for D in slicers}}
+  "{{D['PERS_BACKEND_TYPE']}}",
+{{endfor}}
+]
+
+{{for D in slicers}}
+
+#------------------------------------------------------------------------------
+cdef class {{D['PYTHON_TYPE']}}:
+    cdef {{D['C_TEMPLATE_TYPE']}} truc
+    cdef public vector[vector[double]] filtration_grid
+    cdef public int minpres_degree ## TODO : maybe change directly the degree in the minpres ?
+
+    def __repr__(self):
+        return f"slicer[backend={self.pers_backend},dtype={np.dtype(self.dtype).name},num_param={self.num_parameters},vineyard={self.is_vine},kcritical={self.is_kcritical},is_squeezed={self.is_squeezed},is_minpres={self.is_minpres},max_dim={self.dimension}]" 
+
+    @property
+    def is_squeezed(self)->bool:
+        return self.filtration_grid.size() > 0 and self.filtration_grid[0].size() > 0
+    @property
+    def is_minpres(self)->bool:
+        return self.minpres_degree>=0
+    @staticmethod
+    def _inf_value():
+        return np.asarray(np.inf,dtype={{D['PY_VALUE_TYPE']}}) if issubclass({{D['PY_VALUE_TYPE']}},np.floating) else np.iinfo({{D['PY_VALUE_TYPE']}}).max
+
+    def get_ptr(self):
+        """
+        Returns a pointer to the underlying C++ slicer.
+        """
+        return <intptr_t>(&self.truc)
+    def _from_ptr(self, intptr_t slicer_ptr):
+        self.truc = dereference(<{{D['C_TEMPLATE_TYPE']}}*>(slicer_ptr))
+        return self
+
+    {{if D['IS_SIMPLICIAL']}}
+    def __init__(self, st=None):
+        """
+        Constructs a slicer from a simplex tree.
+        """
+        pass
+    def __cinit__(self, st=None):
+        if st is None:
+            return
+        cdef intptr_t ptr = st.thisptr
+        cdef Simplex_tree_multi_interface[{{D['FILTRATION_TYPE']}},{{D['C_VALUE_TYPE']}}]* st_ptr = <Simplex_tree_multi_interface[{{D['FILTRATION_TYPE']}},{{D['C_VALUE_TYPE']}}]*>(ptr)
+        self.truc = {{D['C_TEMPLATE_TYPE']}}(st_ptr)
+        self.minpres_degree = -1
+    {{elif D['IS_KCRITICAL']}}
+    def __init__(self, generator_maps=[], generator_dimensions=[], filtration_values=[]):        
+        """
+        Constructs a slicer from
+         - scc-like blocks 
+         or 
+         - generator maps (Iterable of list of ints)
+         - generator dimensions (Iterable of int)
+         - filtration values (Iterable of filtration values)
+        """
+        pass
+    def __cinit__(self, generator_maps=[], generator_dimensions=[], filtration_values=[]):
+        """
+        Cython constructor
+        """
+        if len(generator_maps)>0 and len(generator_dimensions) == 0 and len(filtration_values) == 0:
+          from multipers._slicer_meta import _blocks2boundary_dimension_grades
+          generator_maps, generator_dimensions, filtration_values = _blocks2boundary_dimension_grades(
+                  generator_maps,
+                  inplace=False,
+              )
+        cdef uint32_t num_generators = len(generator_maps)
+        cdef vector[vector[uint32_t]] c_generator_maps
+        cdef vector[Multi_critical_filtration[{{D['C_VALUE_TYPE']}}]] c_filtration_values
+        for stuff in generator_maps:
+            c_generator_maps.push_back(<vector[uint32_t]>(stuff))
+        cdef Multi_critical_filtration[{{D['C_VALUE_TYPE']}}] cf
+        cdef One_critical_filtration[{{D['C_VALUE_TYPE']}}] inf
+        inf[0] = -inf[0]
+        cdef {{D['C_VALUE_TYPE']}}[:,:] F_view 
+        for F in filtration_values:
+            cf.push_to_least_common_upper_bound(inf)
+            F_view = np.asarray(F, dtype = {{D['PY_VALUE_TYPE']}} )
+            for i in range(F_view.shape[0]):
+                cf.add_generator(_py21c_{{D['SHORT_VALUE_TYPE']}}(F_view[i]))
+            c_filtration_values.push_back(cf)
+        cdef vector[int] c_generator_dimensions = generator_dimensions
+        assert num_generators == c_generator_maps.size() == c_filtration_values.size(), "Invalid input, shape do not coincide."
+        self.truc = {{D['C_TEMPLATE_TYPE']}}(c_generator_maps,c_generator_dimensions, c_filtration_values)
+        self.minpres_degree = -1
+    {{else}}
+    def __init__(self, generator_maps=[], generator_dimensions=[], filtration_values=[]):        
+        """
+        Constructs a slicer from
+         - generator maps (Iterable of list of ints)
+         - generator dimensions (Iterable of int)
+         - filtration values (Iterable of filtration values)
+        """
+        pass
+    @cython.boundscheck(False)  
+    @cython.wraparound(False)
+    def __cinit__(self, generator_maps=[], generator_dimensions=[], filtration_values=[]):
+        """
+        Cython constructor
+        """
+        cdef uint32_t num_generators = len(generator_maps)
+        filtration_values = np.asarray(filtration_values, dtype = {{D['PY_VALUE_TYPE']}} )
+        assert len(filtration_values) == num_generators, f"Invalid input, shape do not coicide. Got sizes {num_generators,len(generator_dimensions),len(filtration_values)}."
+        cdef vector[vector[uint32_t]] c_generator_maps
+        cdef vector[One_critical_filtration[{{D['C_VALUE_TYPE']}}]] c_filtration_values
+
+        c_generator_maps.resize(num_generators)
+        c_filtration_values.resize(num_generators)
+        for i in range(num_generators):
+            c_generator_maps[i]    = <vector[uint32_t]>(generator_maps[i])
+            c_filtration_values[i] = _py21c_{{D['SHORT_VALUE_TYPE']}}(filtration_values[i])
+        cdef vector[int] c_generator_dimensions = generator_dimensions
+
+        assert num_generators == c_generator_maps.size() == c_filtration_values.size() == c_generator_dimensions.size(), "Invalid input, shape do not coincide."
+        self.truc = {{D['C_TEMPLATE_TYPE']}}(c_generator_maps,c_generator_dimensions, c_filtration_values)
+        self.minpres_degree = -1
+    {{endif}}
+    
+    def copy(self)->{{D['PYTHON_TYPE']}}:
+        """
+        Returns a copy of the slicer.
+        """
+        copy_ = {{D['PYTHON_TYPE']}}()
+        copy_.truc = self.truc
+        copy_.minpres_degree = self.minpres_degree
+        copy_.filtration_grid = self.filtration_grid
+        return copy_
+    def get_barcode(self):
+        """
+        Returns the current barcode.
+        """
+        return self.truc.get_barcode()
+    def push_to_line(self, basepoint, direction=None)->{{D['PYTHON_TYPE']}}:
+        """
+        Pushes the current slicer to the line defined by a basepoint and an optional direction.
+        If the direction is not provided, it is assumed to be diagonal.
+        """
+        basepoint = np.asarray(basepoint, dtype = {{D['PY_VALUE_TYPE']}})
+        cdef Line[{{D['C_VALUE_TYPE']}}] line
+        if direction is None:
+            line = Line[{{D['C_VALUE_TYPE']}}](_py21c_{{D['SHORT_VALUE_TYPE']}}(basepoint))
+        else:
+            direction = np.asarray(direction, dtype = {{D['PY_VALUE_TYPE']}})
+            line = Line[{{D['C_VALUE_TYPE']}}](_py21c_{{D['SHORT_VALUE_TYPE']}}(basepoint),_py21c_{{D['SHORT_VALUE_TYPE']}}(direction))
+        self.truc.push_to(line)
+        return self
+
+    @staticmethod
+    cdef _threshold_bcs(vector[vector[pair[{{D['C_VALUE_TYPE']}}, {{D['C_VALUE_TYPE']}}]]] bcs):
+        return tuple(np.fromiter((a for a in stuff if a.first < {{D['PYTHON_TYPE']}}._inf_value()),  dtype=np.dtype(({{D['PY_VALUE_TYPE']}},2))) for stuff in bcs)
+    @staticmethod
+    def _bc_to_full(bcs, basepoint, direction=None):
+        # i, (b sv d), coords 
+        basepoint = np.asarray(basepoint)[None,None,:]
+        direction = 1 if direction is None else np.asarray(direction)[None,None,:]
+        return tuple(bc[:,:,None]*direction + basepoint for bc in bcs)
+
+    def persistence_on_line(self,basepoint,direction=None, bool keep_inf=True, bool full=False):
+        """
+        Computes the persistence on a line L defined by 
+         - a basepoint (num_parameters,) array
+         - an optional direction (num_parameters,) array
+        
+        Warning: This is not parallelizable. Use `persitence_on_lines`.
+        """
+        self.push_to_line(basepoint,direction)
+        self.truc.compute_persistence()
+        if keep_inf:
+            bcs = tuple(np.asarray(stuff, dtype = {{D['PY_VALUE_TYPE']}}) for stuff in self.truc.get_barcode())
+        else:
+            bcs = {{D['PYTHON_TYPE']}}._threshold_bcs(self.truc.get_barcode())
+
+        if full:
+            bcs = {{D['PYTHON_TYPE']}}._bc_to_full(bcs, basepoint, direction)
+        return bcs
+
+    def persistence_on_lines(self, basepoints=None, directions=None, bool keep_inf=True, bool full=False):
+        """
+        Same as `persistence_on_line`, but with vineyards operation between
+        lines if `self.is_vine`, and in parallel otherwise.
+        """
+        cdef vector[vector[{{D['C_VALUE_TYPE']}}]] c_basepoints
+        cdef vector[pair[vector[{{D['C_VALUE_TYPE']}}], vector[{{D['C_VALUE_TYPE']}}]]] c_truc
+        cdef vector[vector[vector[pair[{{D['C_VALUE_TYPE']}}, {{D['C_VALUE_TYPE']}}]]]] c_out 
+        if directions is None:
+            c_basepoints = basepoints
+            with nogil:
+                c_out = self.truc.persistence_on_lines(c_basepoints)
+        else:
+            c_truc = zip(basepoints,directions)
+            with nogil:
+                c_out = self.truc.persistence_on_lines(c_truc)
+        cdef int num_bc = c_basepoints.size()
+        
+        if keep_inf:
+            out = tuple(tuple(np.asarray(y, dtype = {{D['PY_VALUE_TYPE']}}) for y in x) for x in c_out)
+        else:
+            out = tuple({{D['PYTHON_TYPE']}}._threshold_bcs(x) for x in c_out)
+
+        if full:
+            _dirs = [None]*len(basepoints) if directions is None else directions 
+            out = tuple({{D['PYTHON_TYPE']}}._bc_to_full(bcs, bp, dir) for bcs, bp, dir in zip(out,basepoints,_dirs))
+        return out
+
+
+    def __getstate__(self):
+        return (
+            self.get_boundaries(),
+            self.get_dimensions(),
+            self.get_filtrations(),
+            tuple(np.array(f) for f in self.filtration_grid),
+            self.minpres_degree,
+        )
+    def __setstate__(self, tuple dump):
+        B, D, F, filtration_grid, minpres_degree = dump
+        copy = {{D['PYTHON_TYPE']}}(B,D,F)
+        self.truc = copy.truc
+        self.minpres_degree= minpres_degree
+        self.filtration_grid=filtration_grid
+
+    def __eq__(self, other):
+        ## True if they rpz the same complex (no reordering allowed yet), 
+        # i.e., ignores minpres, squeeze
+        if other.is_squeezed:
+            return self == other.unsqueeze()
+        if self.is_squeezed:
+            return self.unsqueeze() == other
+        return (
+                np.array_equal(self.get_dimensions(), other.get_dimensions())
+                and
+                self.get_boundaries() == other.get_boundaries() 
+                and 
+                ## Kcritical are sorted + filtrationvalues is a dump. 
+                # for non kcritical there is an unnecessary copy though
+                np.array_equal(self.get_filtrations_values(), other.get_filtrations_values())
+            )
+
+
+
+    def compute_persistence(self,one_filtration=None, bool ignore_infinite_filtration_values = True)->{{D['PYTHON_TYPE']}}:
+        """
+        Computes the current persistence, or the persistence
+        given by the filtration one_filtration (num_generators,).
+        """
+        if one_filtration is not None:
+            self.truc.set_one_filtration(one_filtration)
+        # TODO: Later
+        # if len(degrees)>0:
+        #     self.truc.compute_persistence(degrees)
+        # else:
+        #     self.truc.compute_persistence()
+        self.truc.compute_persistence(ignore_infinite_filtration_values)
+        return self
+        # return self.truc.get_barcode()
+    def get_barcode(self):
+        """
+        Returns the barcode of the current 1d-persistence.
+        """
+        return self.truc.get_barcode()
+    def sliced_filtration(self,basepoint, direction=None):
+        """
+        Computes the filtration on a line L defined by 
+         - a basepoint (num_parameters,) array
+         - an optional direction (num_parameters,) array
+        """
+        self.push_to_line(basepoint,direction)
+        return np.asarray(self.truc.get_one_filtration())
+    def __len__(self):
+        return self.truc.num_generators()
+    @property
+    def num_generators(self):
+        return self.truc.num_generators()
+    @property
+    def num_parameters(self):
+        return self.truc.num_parameters()
+    @property
+    def info(self):
+        print(self.truc.to_str().decode())
+    def filtration_bounds(self) -> np.ndarray:
+        """
+        Computes the bounding box of the current multifiltration.
+        """
+        cdef pair[One_critical_filtration[{{D['C_VALUE_TYPE']}}],One_critical_filtration[{{D['C_VALUE_TYPE']}}]] box = self.truc.get_bounding_box()
+        cdef cnp.ndarray[{{D['C_VALUE_TYPE']}}, ndim=1] a = _ff21cview_{{D['SHORT_VALUE_TYPE']}}(&box.first)
+        cdef cnp.ndarray[{{D['C_VALUE_TYPE']}}, ndim=1] b = _ff21cview_{{D['SHORT_VALUE_TYPE']}}(&box.second)
+        return np.asarray([a,b])
+    def get_filtrations_values(self)->np.ndarray:
+        """
+        Returns the current filtration values of the slicer.
+        """
+        cdef vector[One_critical_filtration[{{D['C_VALUE_TYPE']}}]] v = self.truc.get_filtration_values()
+        out = _vff21cview_{{D['SHORT_VALUE_TYPE']}}(v, copy=True, duplicate=self.num_parameters)
+        return np.asarray(out)
+    def get_filtration_grid(self,grid_strategy:str="exact", **infer_grid_kwargs):
+        return mpg.compute_grid(
+                self.get_filtrations_values().T,
+                strategy=grid_strategy,
+                **infer_grid_kwargs,
+            )
+    def get_filtrations(self):
+        """
+        Returns a view of the filtration values, as a list of numpy arrays.
+        """
+        {{if D['IS_KCRITICAL']}}
+        return _vff2kcview_{{D['SHORT_VALUE_TYPE']}}(self.truc.get_filtrations(), copy=False, duplicate=self.num_parameters)
+        {{else}}
+        return _vff21cview_{{D['SHORT_VALUE_TYPE']}}(self.truc.get_filtrations(), copy=False, duplicate=self.num_parameters)
+        {{endif}}
+
+    def get_dimensions(self)-> np.ndarray:
+        """
+        Returns the ordered dimensions of the generators.
+        """
+        return np.asarray(self.truc.get_dimensions())
+    @property 
+    def dimension(self)-> int:
+        """
+        Returns the maximum dimension of the complex.
+        """
+        return self.get_dimensions()[-1] if len(self)>0 else -np.inf
+    def prune_above_dimension(self,int max_dimension)->{{D['PYTHON_TYPE']}}:
+        """
+        Prunes the generators above a given dimension.
+        """
+        self.truc.prune_above_dimension(max_dimension)
+        return self
+    def get_boundaries(self)->tuple[tuple]:
+        """
+        Returns the boundaries of the generators.
+        """
+        return tuple(tuple(b) for b in self.truc.get_boundaries())
+    def grid_squeeze(self, filtration_grid=None, grid_strategy="exact", resolution:Optional[int]=None, bool coordinates=True, bool inplace = False, bool force=False)->{{D['PYTHON_TYPE'][:-3]+"i32"}}|{{D['PYTHON_TYPE']}}:
+        """
+        Coarsen the filtration values on a grid. This is necessary to compute some invariants.
+
+        If the filtration grid is not given, it is infered from filtration values,
+        using the :func:`multipers.grids.compute_grid` function, whose args are
+          - grid_strategy:str see `multipers.grids.available_strategies`. Defaults to exact.
+          - resolution:int if strategy is not exact.
+
+         - inplace:bool if true, does the operation inplace, i.e., doesn't return a copy.
+        """
+        if not force and self.is_squeezed:
+            raise ValueError("The slicer seems to be already squeezed. Use force=True to resqueeze.")
+        if filtration_grid is None:
+            filtration_grid = mpg.compute_grid(
+                    self.get_filtrations_values().T,
+                    strategy=grid_strategy,
+                    resolution=resolution)
+        cdef vector[vector[{{D['C_VALUE_TYPE']}}]] grid = filtration_grid
+        if inplace or not coordinates:
+            self.truc.coarsen_on_grid_inplace(grid, coordinates)
+            self.filtration_grid = filtration_grid
+        else:
+          {{if D['COLUMN_TYPE'] is None}}
+          raise ValueError("WIP")
+          {{else}}
+          out = {{D['PYTHON_TYPE'][:-3]+"i32"}}()
+          out.truc = self.truc.coarsen_on_grid(grid)
+          out.filtration_grid = filtration_grid
+          out.minpres_degree = self.minpres_degree
+          return out
+          {{endif}}
+        return self
+    def minpres(self,
+        int degree=-1, 
+        list[int] degrees=[],
+        str backend:Literal["mpfree", "2pac"]="mpfree", 
+        str slicer_backend:Literal["matrix","clement","graph"]="matrix",
+        bool vineyard={{D['IS_VINE']}}, 
+        id :Optional[str] = None,
+        dtype = {{D['PY_VALUE_TYPE']}},
+        **minpres_kwargs
+        )->Slicer_type:
+        """
+        Computes the minimal presentation of the slicer, and returns it as a new slicer.
+        See :func:`multipers.slicer.minimal_presentation`.
+        """
+        new_slicer = minimal_presentation(self, degree=degree, degrees=degrees, backend=backend, slicer_backend=slicer_backend, vineyard=vineyard, id=id, **minpres_kwargs)
+        return new_slicer
+
+    @property
+    def dtype(self)->type:
+      return {{D['PY_VALUE_TYPE']}}
+    @property
+    def col_type(self)->str:
+      return "{{D['COLUMN_TYPE']}}"
+    @property
+    def is_vine(self)->bool:
+      return {{D['IS_VINE']}}
+    @property
+    def is_kcritical(self)->bool:
+      return {{D['IS_KCRITICAL']}}
+
+    @property
+    def pers_backend(self)->str:
+      return "{{D['PERS_BACKEND_TYPE']}}"
+
+    {{if D['IS_VINE']}}
+    def vine_update(self,basepoint,direction=None)->{{D['PYTHON_TYPE']}}:
+        """
+        Updates the barcode, on a line, using the vineyard algorithm.
+        """
+        self.push_to_line(basepoint,direction)
+        self.truc.vineyard_update()
+        return self
+    def get_representative_cycles(self, bool update=True, bool detailed=False):
+        """
+        Returns the representative cycles of the current barcode.
+        Recomputes the generators if update=True
+        """
+        return self.truc.get_representative_cycles(update, detailed)
+    def get_permutation(self):
+        """
+        Returns the current generator permutation (w.r.t. vineyard).
+        """
+        return self.truc.get_current_order()
+    {{endif}}
+
+    {{if not D['IS_KCRITICAL']}}
+    def _build_from_scc_file(self, path:os.PathLike, bool rivet_compatible = False, bool reverse = False, int shift_dimension = 0)->{{D['PYTHON_TYPE']}}:
+        """
+        Builds the slicer from the given scc file. Should be empty before, otherwise will be completely overwritten.
+        """
+        cdef string c_path = path.encode(encoding="utf-8")
+        with nogil:
+          self.truc.build_from_scc_file(c_path, rivet_compatible, reverse, shift_dimension)
+        return self
+
+    def to_scc(
+            self,
+            path:os.PathLike,
+            int num_parameters = -1,
+            int degree = 1,
+            bool rivet_compatible = False,
+            bool ignore_last_generators = False,
+            bool strip_comments = False,
+            bool reverse = False,
+            bool unsqueeze = True,
+        ):
+        """
+        Writes current slicer to a file in scc format.
+        """
+        cdef string c_path = path.encode(encoding="utf-8")
+        if self.is_squeezed and unsqueeze:
+            kwargs = locals()
+            kwargs.pop("self",0)
+            kwargs.pop("c_path",0)
+            self.unsqueeze().to_scc(**kwargs)
+            return
+        with nogil:
+          self.truc.write_to_scc_file(c_path, num_parameters, degree, rivet_compatible, ignore_last_generators, strip_comments, reverse)
+    def unsqueeze(self, grid=None, dtype = np.float64):
+        from multipers.grids import evaluate_in_grid
+        from multipers import Slicer
+        grid = tuple(np.asarray(f) for f in self.filtration_grid) if grid is None else grid
+        new_filtrations = evaluate_in_grid(np.asarray(self.get_filtrations(), dtype=np.int32), grid)
+        new_slicer = {{D['PYTHON_TYPE'][:-3]+"f64"}}(
+            self.get_boundaries(),
+            self.get_dimensions(),
+            new_filtrations,
+        )
+        new_slicer.minpres_degree=self.minpres_degree
+        return new_slicer
+    {{endif}}
+
+
+{{endfor}}
+
+from libcpp.vector cimport vector
+from libcpp.set cimport set as cset
+cdef extern from "gudhi/cubical_to_boundary.h" namespace "":
+  void _to_boundary(const vector[unsigned int]&, vector[vector[unsigned int]]&, vector[int]&) except + nogil
+  void get_vertices(unsigned int, cset[unsigned int]&, const vector[vector[unsigned int]]&) nogil
+
+
+def from_bitmap(image, **slicer_kwargs):
+    from multipers import Slicer
+    image = np.asarray(image)
+    slicer_kwargs["dtype"] = slicer_kwargs.get("dtype", image.dtype)
+    _Slicer = Slicer(return_type_only=True, **slicer_kwargs)
+    cdef vector[unsigned int] img_shape = image.shape[:-1]
+    cdef unsigned int num_parameters = image.shape[-1]
+    cdef vector[vector[unsigned int]] gen_maps
+    cdef vector[int] gen_dims 
+
+    with nogil:
+        _to_boundary(img_shape,gen_maps, gen_dims)
+
+    cdef unsigned int num_gens = gen_dims.size()
+    filtration_values = np.zeros(shape=(num_gens, num_parameters), dtype = np.double) - np.inf
+    cdef double[:,:] F = filtration_values
+    cdef double[:,:] c_img = image.reshape(-1,num_parameters)
+    cdef cset[unsigned int] vertices
+    with nogil:
+        for i in range(num_gens):
+            # with gil:
+            #     print(f"idx {i}:", end="")
+            vertices.clear()
+            get_vertices(i,vertices,gen_maps)
+
+            # with gil:
+            #     print(f"v = {vertices}:", end="")
+            for k in vertices:
+                for j in range(num_parameters):
+                    F[i,j] =  max(F[i,j], c_img[k,j])
+
+            # with gil:
+            #     print(f"F = {np.asarray(F[i])}")
+    slicer = _Slicer(gen_maps, gen_dims, filtration_values)
+    return slicer
+
+def from_function_delaunay(
+    points,
+    grades,
+    int degree=-1,
+    backend: Optional[_valid_pers_backend]=None,
+    vineyard:Optional[bool]=None,
+    dtype=np.float64,
+    bool verbose = False,
+    bool clear = True,
+):
+    """
+    Given points in $\mathbb R^n$ and function grades, compute the function-delaunay
+    bifiltration as a in an scc format, and converts it into a slicer.
+
+    points : (num_pts, n) float array
+    grades : (num_pts,) float array
+    degree (opt) : if given, computes a minimal presentation of this homological degree first
+    backend : slicer backend, e.g. "matrix", "clement"
+    vineyard : bool, use a vineyard-compatible backend
+    """
+    s = multipers.Slicer(None, backend=backend, vineyard=vineyard, dtype=dtype)
+    mio.function_delaunay_presentation_to_slicer(s, points, grades, degree=degree, verbose=verbose,clear=clear)
+    if degree >= 0:
+        s.minpres_degree = degree
+    return s
+
+def slicer2blocks(slicer, int degree = -1, bool reverse=True):
+    """
+    Convert any slicer to the block format a.k.a. scc format for python
+    """
+    dims = slicer.get_dimensions()
+    num_empty_blocks_to_add = 1 if degree == -1 else dims.min()-degree +1
+    _,counts = np.unique(dims, return_counts=True, )
+    indices = np.concatenate([[0],counts], dtype=np.int32).cumsum()
+    filtration_values = slicer.get_filtrations()
+    filtration_values = [filtration_values[indices[i]:indices[i+1]] for i in range(len(indices)-1)]
+    boundaries = slicer.get_boundaries()
+    boundaries = [boundaries[indices[i]:indices[i+1]] for i in range(len(indices)-1)]
+    shift = np.concatenate([[0], indices], dtype=np.int32)
+    boundaries = [tuple(np.asarray(x-s, dtype=np.int32) for x in block)  for s,block in zip(shift,boundaries)]
+    blocks = [tuple((f,tuple(b))) for b,f in zip(boundaries,filtration_values)]
+    blocks = ([(np.empty((0,)),[])]*num_empty_blocks_to_add) + blocks
+    if reverse:
+        blocks.reverse()
+    return blocks
+
+def minimal_presentation(
+        slicer,
+        int degree = -1, 
+        degrees:Iterable[int]=[],
+        str backend:Literal["mpfree", "2pac", ""]="mpfree", 
+        str slicer_backend:Literal["matrix","clement","graph"]="matrix",
+        bool vineyard=True, 
+        id :Optional[str] =None,
+        dtype:type|_valid_dtypes=None,
+        int n_jobs = -1,
+        bool force=False,
+        **minpres_kwargs
+        ):
+    """
+    Computes a minimal presentation of the multifiltered complex given by the slicer,
+    and returns it as a slicer.
+    Backends differents than `mpfree` are unstable.
+    """
+    if is_slicer(slicer) and slicer.is_minpres and not force:
+        from warnings import warn
+        warn(f"The slicer seems to be already reduced, from homology of degree {slicer.minpres_degree}.")
+        return slicer
+    mio._init_external_softwares(requires=[backend])
+    if len(degrees)>0:
+        def todo(int degree):
+            return minimal_presentation(slicer, degree=degree, backend=backend, slicer_backend=slicer_backend, vineyard=vineyard, id=id, **minpres_kwargs)
+        return tuple(
+          Parallel(n_jobs=n_jobs, backend="threading")(delayed(todo)(d) for d in degrees)
+        )
+        # return tuple(minimal_presentation(slicer, degree=d, backend=backend, slicer_backend=slicer_backend, vineyard=vineyard, id=id, **minpres_kwargs) for d in degrees)
+    assert degree>=0, f"Degree not provided."
+    filtration_grid = slicer.filtration_grid if slicer.is_squeezed else None
+    if id is None:
+        id = str(threading.get_native_id())
+    if dtype is None:
+        dtype = slicer.dtype
+    dimension = slicer.dimension - degree # latest  = L-1, which is empty, -1 for degree 0, -2 for degree 1 etc.
+    slicer.to_scc(path=mio.input_path+id, strip_comments=True, degree=degree-1)
+    new_slicer = multipers.Slicer(None,backend=slicer_backend, vineyard=vineyard, dtype=dtype)
+    if backend=="mpfree":
+        shift_dimension=degree-1
+    else:
+        shift_dimension=degree
+    mio.scc_reduce_from_str_to_slicer(path=mio.input_path+id, slicer=new_slicer, dimension=dimension, backend=backend, shift_dimension=shift_dimension, **minpres_kwargs)
+    new_slicer.minpres_degree = degree
+    if filtration_grid is not None:
+      new_slicer.filtration_grid = filtration_grid
+    return new_slicer
+
+
+def to_simplextree(s:Slicer_type, max_dim:int=-1) -> SimplexTreeMulti_type:
+    """
+    Turns a --simplicial-- slicer into a simplextree.
+
+    Warning: Won't work for non-simplicial complexes, 
+    i.e., complexes $K$ not satisfying 
+    $\forall \sigma \in K,\, \mathrm{dim}(\sigma) = |\partial \sigma|-1$
+    """
+    dims = s.get_dimensions()
+    assert np.all(dims[:-1] <= dims[1:]), "Dims is not sorted."
+    idx = np.searchsorted(dims, np.unique(dims))
+    idx = np.concatenate([idx, [dims.shape[0]]])
+    if max_dim>=0:
+        idx = idx[:max_dim+2]
+
+    cdef vector[vector[int]] boundaries_ = s.get_boundaries()
+    cdef int a
+    cdef int b
+    if len(idx)>2:
+        a = idx[2]
+        b = idx[-1]
+        for i in range(a, b):
+            boundaries_[i] = np.unique(np.concatenate([boundaries_[k] for k in boundaries_[i]]))
+    cdef int num_dims = len(idx)-1
+    boundaries = [np.asarray(boundaries_[idx[i]:idx[i+1]], dtype=np.int32).T for i in range(num_dims)]
+    boundaries[0] = np.arange(boundaries[0].shape[1])[None,:]
+    filtrations = s.get_filtrations()
+    num_parameters  = s.num_parameters
+    filtrations=tuple(filtrations[idx[i]:idx[i+1]] for i in range(num_dims)) # TODO : optimize ?
+    st = SimplexTreeMulti(num_parameters = num_parameters, dtype = s.dtype, kcritical=s.is_kcritical)
+    for dim in range(num_dims):
+        if s.is_kcritical: ## TODO : this may be very slow
+          # for b,f in zip(boundaries[i].T,filtrations[i]):
+          #   for g in np.asarray(f):
+          #       st.insert(np.asarray(b, dtype = np.int32),np.asarray(g, dtype=s.dtype))
+          for j in range(boundaries[i].shape[1]):
+              splx = boundaries[i][:,j]
+              for k in range(filtrations[i][j]):
+                st.insert(splx, np.asarray(filtrations[i][j][k], dtype = s.dtype))
+        else:
+          st.insert_batch(np.asarray(boundaries[dim], dtype= np.int32),np.asarray(filtrations[dim], dtype=s.dtype))
+    return st
+
+
+def _is_slicer(object input)->bool:
+    """
+    Checks if the input is a slicer. Equivalent (but faster) to `isinstance(input, multipers.slicer.Slicer_type)`.
+    """
+    return (False 
+    {{for D in slicers}}
+        or isinstance(input, {{D['PYTHON_TYPE']}})
+        {{endfor}}
+      )
+def is_slicer(input, bool allow_minpres=True)->bool:
+    if _is_slicer(input):
+        return True
+    if allow_minpres and isinstance(input, list) or isinstance(input, tuple):
+        if len(input)>0 and all((_is_slicer(s) and s.is_minpres for s in input)):
+            return True
+    return False
+
+
+def to_blocks(input):
+    """
+    Converts input to blocks, if possible.
+    """
+    if is_slicer(input):
+      return slicer2blocks(input)
+    if isinstance(input, list) or isinstance(input, tuple):
+      return input
+    from multipers.simplex_tree_multi import is_simplextree_multi
+    if is_simplextree_multi(input):
+      return input._to_scc()
+    if isinstance(input, str) or isinstance(input, os.PathLike):
+      return mio.scc_parser(input)
+    raise ValueError("Input cannot be converted to blocks.")
+
+
+def get_matrix_slicer(bool is_vineyard, bool is_k_critical, dtype:type|_valid_dtypes, str col, str pers_backend):
+  """
+  Given various parameters, returns the specific slicer type associated with them.
+  """
+  if False:
+      pass
+{{for D in slicers}}
+{{if not D['IS_SIMPLICIAL']}}
+  elif is_vineyard == {{D['IS_VINE']}} and is_k_critical  ==  {{D['IS_KCRITICAL']}} and np.dtype(dtype) is np.dtype({{D['PY_VALUE_TYPE']}}) and col.lower() == "{{D['COLUMN_TYPE']}}".lower() and "{{D['PERS_BACKEND_TYPE']}}".lower() == pers_backend.lower():
+    return {{D['PYTHON_TYPE']}}
+{{endif}}
+{{endfor}}
+  else:
+    raise ValueError(f"Unimplemented combo for {pers_backend} : {is_vineyard=}, {is_k_critical=}, {dtype=}, {col=}")
+
+
+
+
+def _hilbert_signed_measure(slicer, 
+    vector[indices_type] degrees, 
+    bool zero_pad=False, 
+    indices_type n_jobs=0, 
+    bool verbose=False,
+    # bool expand_collapse=False, 
+    # grid_conversion = None,
+    bool ignore_inf = True,
+    ):
+  """
+  Computes the signed measures given by the decomposition of the hilbert function.
+
+  Input
+  -----
+
+   - simplextree:SimplexTreeMulti, the multifiltered simplicial complex
+   - degrees:array-like of ints, the degrees to compute
+   - n_jobs:int, number of jobs. Defaults to #cpu, but when doing parallel computations of signed measures, we recommend setting this to 1.
+   - verbose:bool, prints c++ logs.
+  
+  Output
+  ------
+  
+  `[signed_measure_of_degree for degree in degrees]`
+  with `signed_measure_of_degree` of the form `(dirac location, dirac weights)`.
+  """
+
+  assert slicer.is_squeezed, "Squeeze grid first."
+  if slicer.is_squeezed:
+    grid_shape = np.array([len(f) for f in slicer.filtration_grid])
+  else:
+      grid_shape = (slicer.filtration_bounds()[1]).astype(python_indices_type)+1
+  if zero_pad:
+    for i, _ in enumerate(grid_shape):
+      grid_shape[i] += 1 # adds a 0
+  assert len(grid_shape) == slicer.num_parameters, "Grid shape size has to be the number of parameters."
+  grid_shape_with_degree = np.asarray(np.concatenate([[len(degrees)], grid_shape]), dtype=python_indices_type)
+  container_array = np.ascontiguousarray(np.zeros(grid_shape_with_degree, dtype=python_tensor_dtype).flatten())
+  assert len(container_array) < np.iinfo(np.uint32).max, "Too large container. Raise an issue on github if you encounter this issue. (Due to tensor's operator[])"
+  cdef vector[indices_type] c_grid_shape = grid_shape_with_degree
+  cdef tensor_dtype[::1] container = container_array
+  cdef tensor_dtype* container_ptr = &container[0]
+  cdef signed_measure_type out = _compute_hilbert_sm(slicer, container_ptr, c_grid_shape, degrees,n_jobs, verbose, zero_pad, ignore_inf)
+  pts, weights = np.asarray(out.first, dtype=python_indices_type).reshape(-1, slicer.num_parameters+1), np.asarray(out.second, dtype=python_tensor_dtype)
+  slices = np.concatenate([np.searchsorted(pts[:,0], np.arange(degrees.size())), [pts.shape[0]] ])
+  sms = [
+      (pts[slices[i]:slices[i+1],1:],weights[slices[i]:slices[i+1]])
+      for i in range(slices.shape[0]-1)
+  ]
+  return sms
+
+
+## Rank invariant
+
+
+## TODO : It is not necessary to do the Möbius inversion in python.
+## fill rank in flipped death, then differentiate in cpp, then reflip with numpy.
+def _rank_from_slicer(
+        slicer, 
+        vector[indices_type] degrees,
+        bool verbose=False,
+        indices_type n_jobs=1,
+        bool zero_pad = False,
+        grid_shape=None,
+        bool plot=False,
+        bool return_raw=False,
+        bool ignore_inf = True,
+        ):
+    # cdef intptr_t slicer_ptr = <intptr_t>(slicer.get_ptr())
+    if grid_shape is None:
+        if slicer.is_squeezed:
+          grid_shape = np.array([len(f) for f in slicer.filtration_grid])
+        else:
+            grid_shape = (slicer.filtration_bounds()[1]).astype(python_indices_type)+1
+    grid_shape = np.asarray(grid_shape)
+
+    cdef int num_parameters = len(grid_shape)
+
+    if zero_pad:
+        for i, _ in enumerate(grid_shape):
+            grid_shape[i] += 1 # adds a 0
+        # for i,f in enumerate(grid_conversion):
+        #     grid_conversion[i] = np.concatenate([f, [mass_default[i]]])
+
+    grid_shape_with_degree = np.asarray(np.concatenate([[len(degrees)], grid_shape, grid_shape]), dtype=python_indices_type)
+    container_array = np.ascontiguousarray(np.zeros(grid_shape_with_degree, dtype=python_tensor_dtype).ravel())
+    assert len(container_array) < np.iinfo(python_indices_type).max, "Too large container. Raise an issue on github if you encounter this issue. (Due to tensor's operator[])"
+    cdef vector[indices_type] c_grid_shape = grid_shape_with_degree
+    cdef tensor_dtype[::1] container = container_array
+    cdef tensor_dtype* container_ptr = &container[0]
+
+    ## SLICERS
+    _compute_rank_invariant(slicer, container_ptr, c_grid_shape, degrees, n_jobs, ignore_inf)
+
+    rank = container_array.reshape(grid_shape_with_degree)
+    rank = tuple(rank_decomposition_by_rectangles(rank_of_degree, threshold = zero_pad) for rank_of_degree in rank)
+    if return_raw:
+        return rank
+    out = []
+    def clean_rank(rank_decomposition):
+        (coords, weights) = sparsify(np.ascontiguousarray(rank_decomposition))
+        births = coords[:,:num_parameters]
+        deaths = coords[:,num_parameters:]
+        correct_indices = np.all(births<=deaths, axis=1)
+        coords = coords[correct_indices]
+        weights = weights[correct_indices]
+        return coords, weights
+
+    out = tuple(clean_rank(rank_decomposition) for rank_decomposition in rank)
+    return out