From c1be3fa5a70b8659e1436c6c29b634a6c88133f4 Mon Sep 17 00:00:00 2001 From: "Dr. Kashif Rasul" Date: Sat, 4 Jan 2020 12:41:21 +0100 Subject: [PATCH] typos --- pts/model/deepar/deepar_network.py | 2 +- pts/model/deepvar/deepvar_network.py | 10 +++++----- test/model/test_deepvar.py | 2 +- 3 files changed, 7 insertions(+), 7 deletions(-) diff --git a/pts/model/deepar/deepar_network.py b/pts/model/deepar/deepar_network.py index be5f1c2..766a5ee 100644 --- a/pts/model/deepar/deepar_network.py +++ b/pts/model/deepar/deepar_network.py @@ -283,7 +283,7 @@ class DeepARTrainingNetwork(DeepARNetwork): else observed_values.min(dim=-1, keepdim=False) ) - weighted_loss = weighted_average(loss, loss_weights) + weighted_loss = weighted_average(loss, weights=loss_weights) return weighted_loss, loss diff --git a/pts/model/deepvar/deepvar_network.py b/pts/model/deepvar/deepvar_network.py index 125bd35..cb6051a 100644 --- a/pts/model/deepvar/deepvar_network.py +++ b/pts/model/deepvar/deepvar_network.py @@ -153,7 +153,7 @@ class DeepVARTrainingNetwork(nn.Module): # (batch_size, seq_len, target_dim * embed_dim) repeated_index_embeddings = ( index_embeddings.unsqueeze(1) - .expand(-1, unroll_length, -1) + .expand(-1, unroll_length, -1, -1) .reshape((-1, unroll_length, self.target_dim * self.embed_dim)) ) @@ -263,7 +263,7 @@ class DeepVARTrainingNetwork(nn.Module): # scale shape is (batch_size, 1, target_dim) _, scale = self.scaler( past_target_cdf[:, -self.context_length :, ...], - past_observed_values[:, -self.context_length : ...,], + past_observed_values[:, -self.context_length :, ...], ) outputs, states, lags_scaled, inputs = self.unroll( @@ -401,17 +401,17 @@ class DeepVARTrainingNetwork(nn.Module): # mask the loss at one time step if one or more observations is missing # in the target dimensions (batch_size, subseq_length, 1) - loss_weights = observed_values.min(dim=-1, keepdim=True) + loss_weights,_ = observed_values.min(dim=-1, keepdim=True) # assert_shape(loss_weights, (-1, seq_len, 1)) - loss = weighted_average(x=likelihoods, weights=loss_weights, dim=1) + loss = weighted_average(likelihoods, weights=loss_weights, dim=1) # assert_shape(loss, (-1, -1, 1)) self.distribution = distr - return (loss, likelihoods) + distr_args + return (loss.sum(), likelihoods) + distr_args class DeepVARPredictionNetwork(DeepVARTrainingNetwork): diff --git a/test/model/test_deepvar.py b/test/model/test_deepvar.py index 4d0e7a5..a8a931d 100644 --- a/test/model/test_deepvar.py +++ b/test/model/test_deepvar.py @@ -87,7 +87,7 @@ def test_deepvar( ): estimator = Estimator( - input_size=10, + input_size=44, num_cells=20, num_layers=1, pick_incomplete=True,