Observation

class nasim.envs.observation.Observation(state_shape)

An observation for NASim.

Each observation is a 2D tensor with a row for each host and an additional row containing auxiliary observations. Each host row is a host_vector (for details see HostVector) while the auxiliary row contains non-host specific observations (see Notes section).

…

obs_shape

the shape of the observation

Type:(int, int)
aux_row

the row index for the auxiliary row

Type:int
tensor

2D Numpy array storing the observation

Type:numpy.ndarray

Notes

The auxiliary row is the final row in the observation tensor and has the following features (in order):

  1. Action success - True (1) or False (0)
    indicates whether the action succeeded or failed
  2. Connection error - True (1) or False (0)
    indicates whether there was a connection error or not
  3. Permission error - True (1) or False (0)
    indicates whether there was a permission error or not
  4. Undefined error - True (1) or False (0)
    indicates whether there was an undefined error or not (e.g. failure due to stochastic nature of exploits)

Since the number of features in the auxiliary row is less than the number of features in each host row, the remainder of the row is all zeros.

__init__(state_shape)
Parameters:state_shape ((int, int)) – 2D shape of the state (i.e. num_hosts, host_vector_size)
connection_error

Whether there was a connection error or not

Returns:True if there was a connection error, otherwise False
Return type:bool
get_readable()

Get a human readable version of the observation

Returns:
  • list[dict] – list of host observations as human-readable dictionary
  • dict[str, bool] – auxiliary observation dictionary
numpy()

Get the observation tensor

Returns:the (2D) observation tenser
Return type:numpy.ndarray
numpy_flat()

Get the flattened observation tensor

Returns:the flattened (1D) observation tenser
Return type:numpy.ndarray
permission_error

Whether there was a permission error or not

Returns:True if there was a permission error, otherwise False
Return type:bool
shape()

Get the (2D) shape of the observation

Returns:the 2D shape of the observation
Return type:(int, int)
shape_flat()

Get the flat (1D) shape of the Observation.

Returns:the flattened shape of observation
Return type:(int, )
success

Whether the action succeded or not

Returns:True if the action succeeded, otherwise False
Return type:bool
undefined_error

Whether there was an undefined error or not

Returns:True if there was a undefined error, otherwise False
Return type:bool