Write a dedicated iostream style class for interacting with IO because it's not entirely obvious how we're going to have to deal with input and output. So the goal is that we can write classes that derive from a base I/O stream class to easily configure this later on.
Action items:
I recommend testing this with the network.h class used for TCP sockets in the MVP. The TCP protocol is incredibly simple, 4 bytes which is batch number N, then N Tensors (float) of size ( i believe 3, 128. So total 4 * 3 * 128 bytes ). Then we just send a single byte 0-11 which is argmax of the result for each of the Tensors we process.
Write a dedicated iostream style class for interacting with IO because it's not entirely obvious how we're going to have to deal with input and output. So the goal is that we can write classes that derive from a base I/O stream class to easily configure this later on.
Action items:
I recommend testing this with the
network.hclass used for TCP sockets in the MVP. The TCP protocol is incredibly simple, 4 bytes which is batch number N, then N Tensors (float) of size ( i believe 3, 128. So total 4 * 3 * 128 bytes ). Then we just send a single byte 0-11 which is argmax of the result for each of the Tensors we process.