lenoctambule/autoencoder
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: working implementation of VAE

Browse Source
This commit is contained in:
Lenoctambule
2026-04-05 01:17:51 +02:00
parent 577e679425
commit 5a8fb2c48b
3 changed files with 42 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
layers.py
View File

@@ -29,7 +29,7 @@ class NNLayer:
return self.output
def backprop(self, error: np.ndarray) -> np.ndarray:
error *= self.activation_func.derivative(self.output_linear)
error *= self.activation_func.d(self.output_linear)
ret = self.W @ error
dW = np.outer(self.input, error) * self.lr
dB = error * self.lr
@@ -57,12 +57,15 @@ class SampleLayer:
def forward(self, v: np.ndarray) -> np.ndarray:
self.input = v
mean = self.mean_nn.forward(v)
std = self.std_nn.forward(v)
return np.random.normal(mean, std, 1)
self.mean = self.mean_nn.forward(v)
self.std = self.std_nn.forward(v)
self.eps = np.random.normal(0, 1)
return self.eps * self.std + self.mean
def backprop(self, errors: np.ndarray) -> np.ndarray:
pass
def backprop(self, error: np.ndarray) -> np.ndarray:
mu_error = self.mean_nn.backprop(error)
std_error = self.std_nn.backprop(self.eps * error)
return mu_error + std_error
class DeepNNLayer: