prediction

tft/module.py

@@ -158,7 +158,7 @@ class TemporalFusionEncoder(nn.Module):
             skip = self.skip_proj(skip)
         encodings = self.gate(encodings)
         encodings = self.lnorm(skip + encodings)
-        return encodings, states
+        return encodings
 
 
 class TemporalFusionDecoder(nn.Module):
@@ -509,7 +509,7 @@ class TFTModel(nn.Module):
         c_c = self.state_c(static_var)
         states = [c_h.unsqueeze(0), c_c.unsqueeze(0)]
 
-        enc_out, _ = self.temporal_encoder(past_selection, future_selection, states)
+        enc_out = self.temporal_encoder(past_selection, future_selection, states)
 
         dec_output = self.temporal_decoder(enc_out, static_enrichment)
 
@@ -569,16 +569,27 @@ class TFTModel(nn.Module):
         c_c = self.state_c(static_var)
         states = [c_h.unsqueeze(0), c_c.unsqueeze(0)]
 
-        enc_out, states = self.temporal_encoder(
+        enc_out = self.temporal_encoder(
             past_selection, tgt_input=None, states=states
         )
+        dec_output = self.temporal_decoder(enc_out, static_enrichment)
+        params = self.param_proj(dec_output)
 
-        for k in range(self.prediction_length):
-            import pdb
+        distr = self.output_distribution(params, scale=scale, trailing_n=1)
 
-            pdb.set_trace()
+        next_sample = distr.sample()
+        future_samples = [next_sample]
+        import pdb
+        pdb.set_trace()
 
+        for k in range(1, self.prediction_length):
             # TODO
+            future_target_proj = pass
+            future_time_feat_proj = pass
+
+            future_selection, _ = self.future_selection(
+                [future_target_proj, future_time_feat_proj], static_selection
+            )
             enc_out, states = self.temporal_encoder(
                 past_selection, future_selection, states
             )