def init_concat_lstm()

in src/model.py [0:0]

• 25 lines of code
• 3 McCabe index (conditional complexity)

def init_concat_lstm(model_config, vocab_size):
    dimension_params = {
        "emb_dim":int(model_config.get("emb_dim")),
        "context_dim":int(model_config.get("context_dim",
                      fallback=int(model_config.get("emb_dim")))),
        "hidden_dim":int(model_config.get("hidden_dim")),
        "vocab_size":vocab_size,
    }

    metadata_constructor_params = {
        "md_projection_dim":int(model_config.get("md_projection_dim", fallback="50")),
        "md_dims":[int(x) for x in model_config.get("md_dims").split(',')],
        "md_group_sizes":[int(x) for x in model_config.get("md_group_sizes").split(',')],
        "attention_mechanism":model_config.get("attention_mechanism", fallback=""),
        "query_type":model_config.get("query_type", fallback=""),
        "use_null_token":eval(model_config.get("use_null_token", fallback="False")),
        "hierarchical_attention":eval(model_config.get("hierarchical_attention", fallback="False")),
    }

    layer_params = {
        "n_layers":1, # Fixed in ACL paper
        "use_softmax_adaptation":eval(model_config.get("use_softmax_adaptation", fallback='False')),
        "use_layernorm":eval(model_config.get("use_layernorm", fallback='False')),
        "use_weight_tying":eval(model_config.get("use_weight_tying", fallback='False')),
    }

    model = ConcatLSTM(dimension_params, metadata_constructor_params, layer_params)

    return model