buildMarkovTableGPU function

GPUTensor<Matrix> buildMarkovTableGPU(GPUTensor<Vector> sequence, int order, int numStates, CommandBuffer tape, { GPUTensor<Matrix>? outTensor, GPUTensor<Matrix>? probTableTensor, })

Constructs a Markov chain probability transition table from a given state sequence. Calculates the frequencies of state transitions based on the specified Markov order and numStates, and normalizes these counts to output a probability matrix of shape [numStates^order, numStates]. This Table does not have a gradient calculation. The operations are appended to tape for execution.

Implementation

GPUTensor<Matrix> buildMarkovTableGPU(
    GPUTensor<Vector> sequence, int order, int numStates, CommandBuffer tape,
    {GPUTensor<Matrix>? outTensor, GPUTensor<Matrix>? probTableTensor}) {
  int numHistories = math.pow(numStates, order).toInt();

  // 1. Allocate VRAM for count and probability tables
  GPUTensor<Matrix> countTable =
      outTensor ?? GPUTensor<Matrix>.empty(<int>[numHistories, numStates]);
  GPUTensor<Matrix> probTable = probTableTensor ??
      GPUTensor<Matrix>.empty(<int>[numHistories, numStates]);
  probTable.subMap["countTable"] = countTable;
  // 2. Zero out the count table (OP_FILL)
  tape.putInt(OP_FILL);
  tape.putString(countTable.id);
  tape.putFloat(0.0);

  // 3. Count Transitions
  tape.putInt(OP_MARKOV_COUNT);
  tape.putString(sequence.id);
  tape.putString(countTable.id);
  tape.putInt(order);
  tape.putInt(numStates);

  // 4. Normalize Counts to Probabilities
  tape.putInt(OP_MARKOV_NORMALIZE);
  tape.putString(countTable.id);
  tape.putString(probTable.id);
  tape.putInt(numHistories);
  tape.putInt(numStates);

  probTable.creator = GPUNode(
      <GPUTensor>[sequence], (CommandBuffer bTape) {}, // Empty backward pass
      opName: 'buildMarkovTableGPU');

  return probTable;
}