Add some minor-impact utilities, rework types in a backward-compatibl…

…e way.

Signed-off-by: Emanuele Ballarin <[email protected]>

ebtorch/nn/__init__.py

@@ -148,8 +148,12 @@
 del tableau10_cycler
 del petroff_2021_cmap
 del tableau10_cmap
+del variadic_attrs
 del set_petroff_2021_colors
 del set_tableau10_colors
+del custom_plot_setup
+del plot_out
+del fromcache
 del repr_sizes_flat_adapter
 del repr_fx_flat_adapter
 del act_auto_broadcast

ebtorch/nn/utils/__init__.py

@@ -16,7 +16,9 @@
 from .actab import broadcast_in_dim
 from .adverutils import AdverApply
 from .adverutils import TA2ATAdapter
+from .attrsutils import variadic_attrs
 from .autoclip import AutoClipper
+from .cacher import fromcache
 from .csttyping import actvt
 from .csttyping import numlike
 from .csttyping import realnum
@@ -49,6 +51,8 @@
 from .patches import patchify_2d
 from .patches import patchify_batch
 from .patches import patchify_dataset
+from .plotting import custom_plot_setup
+from .plotting import plot_out
 from .reprutils import gather_model_repr
 from .reprutils import model_reqgrad
 from .reprutils import model_reqgrad_
@@ -69,5 +73,8 @@
 del reprutils
 del filtermanip
 del palettes
+del cacher
+del plotting
+del attrsutils
 del mapply
 del csttyping

ebtorch/nn/utils/attrsutils.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# ~~ Imports ~~ ────────────────────────────────────────────────────────────────
+from collections.abc import Iterable
+from functools import partial
+from typing import List
+from typing import Optional
+from typing import Tuple
+from typing import Union
+
+# ~~ Exports ~~ ────────────────────────────────────────────────────────────────
+__all__: List[str] = ["variadic_attrs"]
+
+
+# ~~ Utilities ~~ ──────────────────────────────────────────────────────────────
+def _str_to_bool(s: str, onesym: bool = False) -> bool:
+    osl: List[str] = ["t", "y", "1"]
+    if onesym:
+        return s.lower() in osl
+    return s.lower() in (osl + ["true", "yes"])
+
+
+def _any_to_bool(x, onesym: bool = False) -> bool:
+    if isinstance(x, str):
+        return _str_to_bool(x, onesym)
+    return bool(x)
+
+
+def _str_to_booltuple(s: str, sep: Optional[str] = None) -> Tuple[bool, ...]:
+    if sep is not None:
+        return tuple(map(_str_to_bool, s.split(sep)))
+    return tuple(map(partial(_str_to_bool, onesym=True), [*s]))
+
+
+def _any_to_booltuple(
+    x: Union[str, Iterable[Union[str, bool]]], sep: Optional[str] = None
+) -> Tuple[bool, ...]:
+    if isinstance(x, str):
+        return _str_to_booltuple(x, sep)
+    return tuple(map(_any_to_bool, x))
+
+
+def variadic_attrs(
+    selfobj,
+    varsel: Optional[Iterable[Union[str, bool]]] = None,
+    insep: Optional[str] = None,
+    outsep: str = "_",
+):
+    odict: dict = selfobj.__getstate__()
+    odkeys: Tuple[str, ...] = tuple(odict.keys())
+    lodk: int = len(odkeys)
+    varsel: Iterable = varsel if varsel is not None else ([True] * lodk)
+    bvsel: Tuple[bool, ...] = _any_to_booltuple(varsel, insep)
+    strtuple: Tuple[str, ...] = tuple(
+        str(odict[odkeys[i]]) if bvsel[i] else "" for i in range(lodk)
+    )
+    return (outsep.join(strtuple)).strip().strip(outsep)

ebtorch/nn/utils/cacher.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# ~~ Imports ~~ ────────────────────────────────────────────────────────────────
+from collections.abc import Callable
+from typing import List
+from typing import Optional
+from typing import Tuple
+
+# ~~ Exports ~~ ────────────────────────────────────────────────────────────────
+__all__: List[str] = ["fromcache"]
+
+
+# ~~ Utilities ~~ ──────────────────────────────────────────────────────────────
+def _normargs(args, kwargs, kwpos: Tuple[str, ...], kwdef: tuple) -> tuple:
+    largs: int = len(args)
+    n_args: list = list(args) + [None] * (len(kwpos) - largs)
+    for i, argname in enumerate(kwpos[largs:], start=largs):
+        if argname in kwargs:
+            n_args[i] = kwargs[argname]
+        else:
+            n_args[i] = kwdef[i]
+    return tuple(n_args)
+
+
+def _args2keyer(kwpos: Tuple[str, ...], kwdef: tuple) -> Callable:
+    def _args2key(*args, **kwargs) -> tuple:
+        return _normargs(args, kwargs, kwpos, kwdef)
+
+    return _args2key
+
+
+def _retlookup(key: tuple, dictionary: dict) -> Optional:
+    return dictionary[key] if key in dictionary else None
+
+
+# ~~ Cache retrieval function ~~ ───────────────────────────────────────────────
+def fromcache(
+    func: Callable,
+    *,
+    kwpos: Tuple[str, ...],
+    kwdef: tuple,
+    cache: dict,
+    updateable: bool = True,
+) -> Callable:
+    def _cached_func(*args, **kwargs):
+        key: tuple = _args2keyer(kwpos, kwdef)(*args, **kwargs)
+        if (rlk := _retlookup(key, cache)) is not None:
+            return rlk
+        else:
+            if updateable:
+                funcret = func(*args, **kwargs)
+                cache[key] = funcret
+                return funcret
+            else:
+                raise ValueError("Cache miss, but cache is not updateable.")
+
+    return _cached_func

ebtorch/nn/utils/plotting.py

@@ -0,0 +1,34 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# ──────────────────────────────────────────────────────────────────────────────
+from typing import List
+from typing import Optional
+
+import matplotlib.pyplot as plt
+
+# ──────────────────────────────────────────────────────────────────────────────
+__all__: List[str] = ["custom_plot_setup", "plot_out"]
+# ──────────────────────────────────────────────────────────────────────────────
+
+
+def custom_plot_setup() -> None:
+    plt.rcParams["text.usetex"] = True
+    plt.style.use("ggplot")
+    plt.rcParams["axes.facecolor"] = "white"
+    plt.rcParams["axes.edgecolor"] = "black"
+    plt.rcParams["axes.spines.top"] = False
+    plt.rcParams["axes.spines.right"] = False
+    plt.rcParams["xtick.color"] = "black"
+    plt.rcParams["ytick.color"] = "black"
+    plt.rcParams["axes.labelcolor"] = "black"
+    plt.rcParams["grid.color"] = "gainsboro"
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+
+
+def plot_out(savepath: Optional[str] = None) -> None:
+    if savepath:
+        plt.savefig(savepath, dpi=400, bbox_inches="tight")
+    else:
+        plt.show()

ebtorch/typing/customtypes.py

@@ -5,13 +5,15 @@
 from typing import List
 from typing import Union
 
+import numpy as np
 import torch
 from torch import Tensor
 
 # ──────────────────────────────────────────────────────────────────────────────
-__all__: List[str] = ["realnum", "strdev", "numlike"]
+__all__: List[str] = ["realnum", "strdev", "numlike", "tensorlike"]
 # ──────────────────────────────────────────────────────────────────────────────
 realnum = Union[int, float]
 strdev = Union[str, torch.device]
-numlike = Union[realnum, Tensor]
+tensorlike = Union[Tensor, np.ndarray]
+numlike = Union[realnum, tensorlike]
 # ──────────────────────────────────────────────────────────────────────────────

setup.py

@@ -26,7 +26,7 @@ def read(fname):
 
 setup(
     name=PACKAGENAME,
-    version="0.28.2",
+    version="0.28.3",
     author="Emanuele Ballarin",
     author_email="[email protected]",
     url="https://github.com/emaballarin/ebtorch",