Fix bug with lazy data loading, un-implement __len__ on AllennlpLazyDataset

[epwalsh]

This fixes a bug that occurs when using a lazy dataset reader where you'd get a PyTorch warning on every single batch during training. It does so by making AllennlpLazyDataset consistent with torch.data.IterableDataset in that it no longer implements __len__() (previously len() would just return 1).

To reproduce this bug just run allennlp train on a experiment with a lazy dataset reader like this:

{
  "dataset_reader": {
    "type": "squad",
    "lazy": true,
    "token_indexers": {
      "tokens": {
        "type": "single_id",
        "lowercase_tokens": true
      },
      "token_characters": {
        "type": "characters",
        "character_tokenizer": {
          "byte_encoding": "utf-8",
          "start_tokens": [259],
          "end_tokens": [260]
        },
        "min_padding_length": 5
      }
    }
  },
  "train_data_path": "https://allennlp.s3.amazonaws.com/datasets/squad/squad-train-v1.1.json",
  "validation_data_path": "https://allennlp.s3.amazonaws.com/datasets/squad/squad-dev-v1.1.json",
  "model": {
    "type": "bidaf",
    "text_field_embedder": {
      "token_embedders": {
        "tokens": {
          "type": "embedding",
          "pretrained_file": "https://allennlp.s3.amazonaws.com/datasets/glove/glove.6B.100d.txt.gz",
          "embedding_dim": 100,
          "trainable": false
        },
        "token_characters": {
          "type": "character_encoding",
          "embedding": {
            "num_embeddings": 262,
            "embedding_dim": 16
          },
          "encoder": {
            "type": "cnn",
            "embedding_dim": 16,
            "num_filters": 100,
            "ngram_filter_sizes": [5]
          },
          "dropout": 0.2
        }
      }
    },
    "num_highway_layers": 2,
    "phrase_layer": {
      "type": "lstm",
      "bidirectional": true,
      "input_size": 200,
      "hidden_size": 100,
      "num_layers": 1
    },
    "matrix_attention": {
      "type": "linear",
      "combination": "x,y,x*y",
      "tensor_1_dim": 200,
      "tensor_2_dim": 200
    },
    "modeling_layer": {
      "type": "lstm",
      "bidirectional": true,
      "input_size": 800,
      "hidden_size": 100,
      "num_layers": 2,
      "dropout": 0.2,
    },
    "span_end_encoder": {
      "type": "lstm",
      "bidirectional": true,
      "input_size": 1400,
      "hidden_size": 100,
      "num_layers": 1,
    },
    "dropout": 0.2,
  },
  "data_loader": {
    "batch_size": 40,
    "num_workers": 1,
  },
  "trainer": {
    "num_epochs": 20,
    "grad_norm": 5.0,
    // "patience": 10,
    "validation_metric": "+em",
    "cuda_device": 0,
    "learning_rate_scheduler": {
      "type": "reduce_on_plateau",
      "factor": 0.5,
      "mode": "max",
      "patience": 2,
    },
    "optimizer": {
      "type": "adam",
      "betas": [0.9, 0.9],
    },
  },
}

[epwalsh]

Need this so len() is not called.

[epwalsh]

Need this so len() is not called.

[epwalsh]

By the way I tracked the __len__ implementation back to this PR. I think the rational for adding it on lazy datasets is now out-dated: #3700 (comment).

[epwalsh]

I think this should close #4035

[dirkgr]

Why does it default to 1 here? Isn't that a problem when it happens?

[epwalsh]

This gives the same behavior as it was before. I'm not sure if that's an issue.