fix: remove deprecated layers

elasticai/creator/base_modules/arithmetics/binary_arithmetics.py

@@ -6,7 +6,7 @@
     Binarize,
 )
 
-from .arithmetics import Arithmetics
+from ._arithmetics import Arithmetics
 
 
 class BinaryArithmetics(Arithmetics):

elasticai/creator/base_modules/arithmetics/float_arithmetics.py

@@ -2,7 +2,7 @@
 
 import torch
 
-from elasticai.creator.base_modules.arithmetics.arithmetics import Arithmetics
+from elasticai.creator.base_modules.arithmetics._arithmetics import Arithmetics
 from elasticai.creator.base_modules.autograd_functions.round_to_float import (
     RoundToFloat,
 )

elasticai/creator/base_modules/arithmetics/torch_arithmetics.py

@@ -2,7 +2,7 @@
 
 import torch
 
-from .arithmetics import Arithmetics
+from ._arithmetics import Arithmetics
 
 
 class TorchArithmetics(Arithmetics):

elasticai/creator/base_modules/conv1d.py

@@ -1,19 +1,16 @@
-from typing import Any, Protocol
+from typing import Any
 
 from torch import Tensor
 from torch.nn import Conv1d as _Conv1d
 from torch.nn.functional import conv1d
 
-
-class Arithmetics(Protocol):
-    def quantize(self, x: Tensor) -> Tensor:
-        ...
+from elasticai.creator.base_modules.math_operations import Quantize as MathOperations
 
 
 class Conv1d(_Conv1d):
     def __init__(
         self,
-        arithmetics: Arithmetics,
+        arithmetics: MathOperations,
         in_channels: int,
         out_channels: int,
         kernel_size: int | tuple[int],

elasticai/creator/base_modules/linear.py

@@ -1,16 +1,20 @@
-from typing import Any
+from typing import Any, Protocol
 
 import torch
 
-from .arithmetics.arithmetics import Arithmetics
+from elasticai.creator.base_modules.math_operations import Add, MatMul, Quantize
+
+
+class MathOperations(Quantize, Add, MatMul, Protocol):
+    ...
 
 
 class Linear(torch.nn.Linear):
     def __init__(
         self,
         in_features: int,
         out_features: int,
-        arithmetics: Arithmetics,
+        arithmetics: MathOperations,
         bias: bool,
         device: Any = None,
         dtype: Any = None,

elasticai/creator/base_modules/lstm_cell.py

@@ -3,8 +3,7 @@
 
 import torch
 
-from .arithmetics.arithmetics import Arithmetics
-from .linear import Linear
+from .linear import Linear, MathOperations
 
 
 class LSTMCell(torch.nn.Module):
@@ -13,7 +12,7 @@ def __init__(
         input_size: int,
         hidden_size: int,
         bias: bool,
-        arithmetics: Arithmetics,
+        arithmetics: MathOperations,
         sigmoid_factory: Callable[[], torch.nn.Module],
         tanh_factory: Callable[[], torch.nn.Module],
         device: Any = None,

elasticai/creator/base_modules/math_operations.py

@@ -0,0 +1,22 @@
+from abc import abstractmethod
+from typing import Protocol
+
+from torch import Tensor
+
+
+class Quantize(Protocol):
+    @abstractmethod
+    def quantize(self, x: Tensor) -> Tensor:
+        ...
+
+
+class MatMul(Protocol):
+    @abstractmethod
+    def matmul(self, a: Tensor, b: Tensor) -> Tensor:
+        ...
+
+
+class Add(Protocol):
+    @abstractmethod
+    def add(self, a: Tensor, b: Tensor) -> Tensor:
+        ...

elasticai/creator/base_modules/silu.py → elasticai/creator/base_modules/siluwithtrainablescalebeta.py

@@ -1,10 +1,10 @@
 import torch
 
-from elasticai.creator.base_modules.arithmetics.arithmetics import Arithmetics
+from elasticai.creator.base_modules.math_operations import QuantizeOperation
 
 
-class SiLU(torch.nn.SiLU):
-    def __init__(self, arithmetics: Arithmetics) -> None:
+class SiLUWithTrainableScaleBeta(torch.nn.SiLU):
+    def __init__(self, arithmetics: QuantizeOperation) -> None:
         super().__init__(inplace=False)
         self._arithmetics = arithmetics
         self.scale = torch.nn.Parameter(torch.ones(1, requires_grad=True))

elasticai/creator/nn/__init__.py

@@ -1,8 +0,0 @@
-from .conv1d import FPBatchNormedConv1d, FPConv1d
-from .hard_sigmoid import FPHardSigmoid
-from .hard_tanh import FPHardTanh
-from .identity import BufferedIdentity, BufferlessIdentity
-from .linear import FPBatchNormedLinear, FPLinear
-from .precomputed import FPSigmoid, FPTanh
-from .relu import FPReLU
-from .sequential import Sequential

elasticai/creator/nn/fixed_point/_math_operations.py

@@ -0,0 +1,31 @@
+from typing import cast
+
+import torch
+from fixed_point._round_to_fixed_point import RoundToFixedPoint
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
+
+from elasticai.creator.base_modules.conv1d import MathOperations as Conv1dOps
+from elasticai.creator.base_modules.linear import MathOperations as LinearOps
+from elasticai.creator.base_modules.lstm_cell import MathOperations as LSTMOps
+
+
+class MathOperations(LinearOps, LSTMOps, Conv1dOps):
+    def __init__(self, config: FixedPointConfig) -> None:
+        self.config = config
+
+    def quantize(self, a: torch.Tensor) -> torch.Tensor:
+        return self._round(self._clamp(a))
+
+    def _clamp(self, a: torch.Tensor) -> torch.Tensor:
+        return torch.clamp(
+            a, min=self.config.minimum_as_rational, max=self.config.maximum_as_rational
+        )
+
+    def _round(self, a: torch.Tensor) -> torch.Tensor:
+        return cast(torch.Tensor, RoundToFixedPoint.apply(a, self.config))
+
+    def add(self, a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
+        return self._clamp(a + b)
+
+    def matmul(self, a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
+        return self.quantize(torch.matmul(a, b))

elasticai/creator/base_modules/autograd_functions/round_to_fixed_point.py → elasticai/creator/nn/fixed_point/_round_to_fixed_point.py

@@ -1,10 +1,7 @@
 from typing import Any
 
 import torch
-
-from elasticai.creator.base_modules.two_complement_fixed_point_config import (
-    FixedPointConfig,
-)
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
 
 
 class RoundToFixedPoint(torch.autograd.Function):

elasticai/creator/nn/conv1d/layer.py → elasticai/creator/nn/fixed_point/conv1d/layer.py

@@ -2,14 +2,10 @@
 
 import torch
 import torch.nn
+from fixed_point._math_operations import Operations
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
 
-from elasticai.creator.base_modules.arithmetics.fixed_point_arithmetics import (
-    FixedPointArithmetics,
-)
 from elasticai.creator.base_modules.conv1d import Conv1d
-from elasticai.creator.base_modules.two_complement_fixed_point_config import (
-    FixedPointConfig,
-)
 from elasticai.creator.vhdl.translatable import Translatable
 
 from .design import FPConv1d as FPConv1dDesign
@@ -32,7 +28,7 @@ def __init__(
         self._config = FixedPointConfig(total_bits=total_bits, frac_bits=frac_bits)
         self._signal_length = signal_length
         super().__init__(
-            arithmetics=FixedPointArithmetics(config=self._config),
+            arithmetics=Operations(config=self._config),
             in_channels=in_channels,
             out_channels=out_channels,
             kernel_size=kernel_size,
@@ -106,7 +102,7 @@ def __init__(
     ) -> None:
         super().__init__()
         self._config = FixedPointConfig(total_bits=total_bits, frac_bits=frac_bits)
-        self._arithmetics = FixedPointArithmetics(config=self._config)
+        self._arithmetics = Operations(config=self._config)
         self._signal_length = signal_length
         self._conv1d = Conv1d(
             arithmetics=self._arithmetics,

elasticai/creator/nn/hard_sigmoid/layer.py → elasticai/creator/nn/fixed_point/hard_sigmoid/layer.py

@@ -1,7 +1,6 @@
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
+
 from elasticai.creator.base_modules.hard_sigmoid import HardSigmoid
-from elasticai.creator.base_modules.two_complement_fixed_point_config import (
-    FixedPointConfig,
-)
 from elasticai.creator.vhdl.design.design import Design
 from elasticai.creator.vhdl.translatable import Translatable
 

elasticai/creator/nn/hard_tanh/layer.py → elasticai/creator/nn/fixed_point/hard_tanh/layer.py

@@ -1,7 +1,6 @@
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
+
 from elasticai.creator.base_modules.hard_tanh import HardTanh
-from elasticai.creator.base_modules.two_complement_fixed_point_config import (
-    FixedPointConfig,
-)
 from elasticai.creator.vhdl.design.design import Design
 from elasticai.creator.vhdl.translatable import Translatable
 

elasticai/creator/nn/linear/layer.py → elasticai/creator/nn/fixed_point/linear/layer.py

@@ -1,14 +1,10 @@
 from typing import Any, cast
 
 import torch
+from fixed_point._math_operations import Operations
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
 
-from elasticai.creator.base_modules.arithmetics.fixed_point_arithmetics import (
-    FixedPointArithmetics,
-)
 from elasticai.creator.base_modules.linear import Linear
-from elasticai.creator.base_modules.two_complement_fixed_point_config import (
-    FixedPointConfig,
-)
 from elasticai.creator.vhdl.design.design import Design
 from elasticai.creator.vhdl.translatable import Translatable
 
@@ -29,7 +25,7 @@ def __init__(
         super().__init__(
             in_features=in_features,
             out_features=out_features,
-            arithmetics=FixedPointArithmetics(config=self._config),
+            arithmetics=Operations(config=self._config),
             bias=bias,
             device=device,
         )
@@ -71,7 +67,7 @@ def __init__(
         device: Any = None,
     ) -> None:
         super().__init__()
-        self._arithmetics = FixedPointArithmetics(
+        self._arithmetics = Operations(
             config=FixedPointConfig(total_bits=total_bits, frac_bits=frac_bits)
         )
         self._linear = Linear(

elasticai/creator/nn/precomputed/fp_precomputed_module.py → elasticai/creator/nn/fixed_point/precomputed/fp_precomputed_module.py

@@ -1,16 +1,12 @@
 from typing import cast
 
 import torch
+from fixed_point._math_operations import Operations
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
 
-from elasticai.creator.base_modules.arithmetics.fixed_point_arithmetics import (
-    FixedPointArithmetics,
-)
 from elasticai.creator.base_modules.autograd_functions.identity_step_function import (
     IdentityStepFunction,
 )
-from elasticai.creator.base_modules.two_complement_fixed_point_config import (
-    FixedPointConfig,
-)
 from elasticai.creator.vhdl.design.design import Design
 from elasticai.creator.vhdl.shared_designs.precomputed_scalar_function import (
     PrecomputedScalarFunction,
@@ -30,7 +26,7 @@ def __init__(
         super().__init__()
         self._base_module = base_module
         self._config = FixedPointConfig(total_bits=total_bits, frac_bits=frac_bits)
-        self._arithmetics = FixedPointArithmetics(self._config)
+        self._arithmetics = Operations(self._config)
         self._step_lut = torch.linspace(*sampling_intervall, num_steps)
 
     def forward(self, inputs: torch.Tensor) -> torch.Tensor:

elasticai/creator/nn/precomputed/silu.py → elasticai/creator/nn/fixed_point/precomputed/silu.py

@@ -1,9 +1,8 @@
-from elasticai.creator.base_modules.arithmetics.fixed_point_arithmetics import (
-    FixedPointArithmetics,
-)
-from elasticai.creator.base_modules.silu import SiLU
-from elasticai.creator.base_modules.two_complement_fixed_point_config import (
-    FixedPointConfig,
+from fixed_point._math_operations import Operations
+from fixed_point._two_complement_fixed_point_config import FixedPointConfig
+
+from elasticai.creator.base_modules.siluwithtrainablescalebeta import (
+    SiLUWithTrainableScaleBeta,
 )
 
 from .fp_precomputed_module import FPPrecomputedModule
@@ -18,8 +17,8 @@ def __init__(
         sampling_intervall: tuple[float, float] = (-10, 10),
     ) -> None:
         super().__init__(
-            base_module=SiLU(
-                arithmetics=FixedPointArithmetics(
+            base_module=SiLUWithTrainableScaleBeta(
+                arithmetics=Operations(
                     config=FixedPointConfig(total_bits=total_bits, frac_bits=frac_bits)
                 ),
             ),

pyproject.toml

@@ -23,7 +23,6 @@ python-language-server = "^0.36.2"
 black = ">21.12"
 prospector = {extras = ["with_mypy"], version = "^1.7.7"}
 pre-commit = "^3.0.0"
-mypy = "^1.0.0"
 wily = "^1.24.0"
 import-linter = "^1.7.0"
 impulse = "^1.0"