Predicting embeddings for new triples after training a pykeen model using pipeline

Question

I implemented the following code to get the best model using the pipeline in pykeen, and now want to get the embeddings of new triples using best_model. How can this be done?

  model_types = ['TransR', 'CompGCN', 'ConvE']
  losses = ['pointwisehinge', 'pairwisehinge']
  for model_ch in model_types:
    for loss_ch in losses:
      try:
        result = pipeline(
            model=model_ch,
            loss=loss_ch,
            training=training,
            testing=testing,
            validation=validation,
            model_kwargs=dict(embedding_dim=10, random_seed=12),
            optimizer_kwargs=dict(lr=0.025),  
            training_kwargs=dict(num_epochs=100, use_tqdm_batch=False),
        )
        l = result.losses
        if min(l) < least_loss:
          least_loss = min(l)
          best_model = result.model
        print("Model type = " + model_ch)
        print("Loss type = " + loss_ch)`

I tried using best_model.entity_representations[0]._embeddings.weight to extract the weights but do not know how to get the embeddings of new triples.