Delete evaluation/QP/net.py

evaluation/QP/net.py

@@ -1,29 +0,0 @@
-import torch
-from torch import nn, optim
-
-class LossPredictor(nn.Module):
-    def __init__(self, ssl_model, ssl_out_dim, dropout_rate=0.2):
-        super(LossPredictor, self).__init__()
-        self.ssl_model = ssl_model
-        self.ssl_features = ssl_out_dim
-        self.dropout = nn.Dropout(dropout_rate)
-        self.carac_interm = nn.Linear(self.ssl_features,1000) #num dim neurones
-        self.output_layer = nn.Linear(1000, 1) #1 neurones
-
-
-    def forward(self, wav):
-        wav = wav.squeeze(1)
-        res = self.ssl_model(wav, mask=False, features_only=True, layer=None)
-        x = res['x']
-        if 'layer_results' in res:
-            layer_outputs = [layer[0] for layer in res['layer_results']] #12 couches
-
-            concatenated_layers = torch.cat(layer_outputs, dim=-1)
-        
-            x = torch.mean(concatenated_layers, dim=0)
-        
-        x = self.carac_interm(x)
-        
-        x = self.dropout(x) ###Pour le MCDropout pendant train+ pred select candidate
-        x = self.output_layer(x)
-        return x.squeeze(1) #return loss