*   Added **categorical data** processing layers:
    *   `IntegerLookup` & `StringLookup`: build an index of categorical feature
        values
    *   `CategoryEncoding`: turn integer-encoded categories into one-hot,
        multi-hot, or tf-idf encoded representations
    *   `CategoryCrossing`: create new categorical features representing
        co-occurrences of previous categorical feature values

  def maybe_encode_env(env):
    """Encodes unicode in env to str on Windows python 2.x."""
    if not is_windows() or sys.version_info[0] != 2:
      return env
    for k, v in env.items():
      if isinstance(k, unicode):
        k = k.encode('ascii')
      if isinstance(v, unicode):
        v = v.encode('ascii')
      env[k] = v
    return env