kosh/neural-networks-from-scratch
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Changed parameter type from tuple to list in layer.py

Browse Source
This commit is contained in:
Koshin S Hegde 2024-05-09 13:00:54 +05:30
parent 3fe69b3869
commit 98b3b4e18a
1 changed files with 21 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
src/layer.py
View File

@ -3,8 +3,11 @@ from abc import ABC, abstractmethod
class Layer(ABC): class Layer(ABC):
_cached_inputs: np.ndarray
@abstractmethod @abstractmethod
def forward(self, x: np.ndarray) -> np.ndarray: def forward(self, x: np.ndarray) -> np.ndarray:
self._cached_inputs = x
""" """
x = inputs x = inputs
if should_cache = True, if should_cache = True,
@ -14,7 +17,7 @@ class Layer(ABC):
@property @property
@abstractmethod @abstractmethod
def parameters(self) -> tuple[np.ndarray, ...]: def parameters(self) -> list[np.ndarray]:
""" """
Returns the different parameters. Returns the different parameters.
The order is defined as per the sub class's convinience The order is defined as per the sub class's convinience
@ -22,17 +25,17 @@ class Layer(ABC):
@parameters.setter @parameters.setter
@abstractmethod @abstractmethod
def parameters(self, parameters: tuple[np.ndarray, ...]) -> None: def parameters(self, parameters: list[np.ndarray]) -> None:
""" """
Write to parameters property Write to parameters property
""" """
@abstractmethod @abstractmethod
def d_output_wrt_parameters(self, inputs: np.ndarray) -> tuple[np.ndarray, ...]: def d_output_wrt_parameters(self) -> list[np.ndarray]:
pass pass
@abstractmethod @abstractmethod
def d_output_wrt_inputs(self, x: np.ndarray) -> np.ndarray: def d_output_wrt_inputs(self) -> np.ndarray:
pass pass
@ -46,25 +49,26 @@ class Dense(Layer):
self.__b = np.random.random((output_size)) self.__b = np.random.random((output_size))
def forward(self, x: np.ndarray) -> np.ndarray: def forward(self, x: np.ndarray) -> np.ndarray:
super().forward(x)
return np.dot(self.__w, x.T).T + self.__b return np.dot(self.__w, x.T).T + self.__b
@property @property
def parameters(self) -> tuple[np.ndarray, np.ndarray]: def parameters(self) -> list[np.ndarray]:
return (self.__w, self.__b) return [self.__w, self.__b]
@parameters.setter @parameters.setter
def parameters(self, parameters: tuple[np.ndarray, ...]) -> None: def parameters(self, parameters: list[np.ndarray]) -> None:
self.__w = parameters[0] self.__w = parameters[0]
self.__b = parameters[1] self.__b = parameters[1]
def d_output_wrt_parameters(self, inputs: np.ndarray) -> tuple[np.ndarray, np.ndarray]: def d_output_wrt_parameters(self) -> list[np.ndarray]:
""" """
d_out_wrt_w = input d_out_wrt_w = input
d_out_wrt_b = output d_out_wrt_b = output
""" """
return (inputs, np.array([1])) return [self._cached_inputs, np.array([1])]
def d_output_wrt_inputs(self, x: np.ndarray) -> np.ndarray: def d_output_wrt_inputs(self) -> np.ndarray:
return self.__w return self.__w
@ -73,15 +77,15 @@ class ReLU(Layer):
return (x > 0) * x return (x > 0) * x
@property @property
def parameters(self) -> tuple[np.ndarray, ...]: def parameters(self) -> list[np.ndarray]:
return () return []
@parameters.setter @parameters.setter
def parameters(self, parameters: tuple[np.ndarray, ...]) -> None: def parameters(self, parameters: list[np.ndarray]) -> None:
return return
def d_output_wrt_parameters(self, inputs: np.ndarray) -> tuple[np.ndarray, ...]: def d_output_wrt_parameters(self) -> list[np.ndarray]:
return () return []
def d_output_wrt_inputs(self, x: np.ndarray) -> np.ndarray: def d_output_wrt_inputs(self) -> np.ndarray:
return (x > 0) * 1.0 return (self._cached_inputs > 0) * 1.0