foxhunt/infra/scripts/train.sh

#!/usr/bin/env bash
set -euo pipefail

# ---------------------------------------------------------------------------
# train.sh - Training orchestrator: submits K8s Jobs to the GPU pool
# ---------------------------------------------------------------------------
# Usage:
#   ./infra/scripts/train.sh <preset> [--model MODEL] [--symbol SYMBOL]
#                                      [--trials N] [--max-steps N] [--timeout N]
#
# Presets:
#   quick-test    - fast sanity check (max-steps=500, requires --model)
#   single-model  - full training run for one model (requires --model)
#   full-ensemble - trains all 10 models sequentially (dqn ppo tft mamba2 tggn tlob liquid kan xlstm diffusion)
#   hyperopt      - hyperparameter optimisation (requires --model, uses --trials)
# ---------------------------------------------------------------------------

# --- Defaults -------------------------------------------------------------
SYMBOL="ES.FUT"
TRIALS=20
MAX_STEPS=2000
TIMEOUT=3600
DATA_DIR="/data/futures-baseline"
NAMESPACE="foxhunt"
IMAGE="rg.fr-par.scw.cloud/foxhunt/training:latest"
MODEL=""
PRESET=""
TIMESTAMP="$(date +%s)"

ALL_MODELS=(dqn ppo tft mamba2 tggn tlob liquid kan xlstm diffusion)

# --- Model-to-binary mapping ----------------------------------------------
declare -A MODEL_BINARY=(
  [dqn]=train_dqn_es_fut
  [ppo]=train_ppo_parquet
  [tft]=train_tft_dbn
  [mamba2]=train_mamba2_dbn
  [tggn]=train_tggn_dbn
  [tlob]=train_tlob
  [liquid]=train_liquid_dbn
  [kan]=train_kan_dbn
  [xlstm]=train_xlstm_dbn
  [diffusion]=train_diffusion_dbn
)

# --- Helpers --------------------------------------------------------------
usage() {
  cat <<'USAGE'
Usage: ./infra/scripts/train.sh <preset> [OPTIONS]

Presets:
  quick-test      Fast sanity check (requires --model, max-steps=500)
  single-model    Full training run (requires --model)
  full-ensemble   Trains all 10 models sequentially (dqn ppo tft mamba2 tggn tlob liquid kan xlstm diffusion)
  hyperopt        Hyperparameter search (requires --model, uses --trials)

Options:
  --model MODEL     Model name: dqn | ppo | tft | mamba2 | tggn | tlob | liquid | kan | xlstm | diffusion
  --symbol SYMBOL   Trading symbol          (default: ES.FUT)
  --trials N        Hyperopt trial count    (default: 20)
  --max-steps N     Max steps per epoch     (default: 2000)
  --timeout N       Job timeout in seconds  (default: 3600)
  -h, --help        Show this help message
USAGE
  exit 0
}

die() { echo "ERROR: $*" >&2; exit 1; }

validate_model() {
  local m="$1"
  [[ -n "${MODEL_BINARY[$m]+x}" ]] || die "Unknown model '${m}'. Valid: ${!MODEL_BINARY[*]}"
}

# --- Parse arguments -------------------------------------------------------
[[ $# -eq 0 ]] && usage

PRESET="$1"; shift

while [[ $# -gt 0 ]]; do
  case "$1" in
    --model)    MODEL="$2";     shift 2 ;;
    --symbol)   SYMBOL="$2";    shift 2 ;;
    --trials)   TRIALS="$2";    shift 2 ;;
    --max-steps) MAX_STEPS="$2"; shift 2 ;;
    --timeout)  TIMEOUT="$2";   shift 2 ;;
    -h|--help)  usage ;;
    *) die "Unknown option: $1" ;;
  esac
done

# --- Preset validation -----------------------------------------------------
case "$PRESET" in
  quick-test)
    [[ -z "$MODEL" ]] && die "quick-test preset requires --model"
    validate_model "$MODEL"
    MAX_STEPS=500
    ;;
  single-model)
    [[ -z "$MODEL" ]] && die "single-model preset requires --model"
    validate_model "$MODEL"
    ;;
  full-ensemble)
    # Trains all models; --model is ignored if provided
    MODEL=""
    ;;
  hyperopt)
    [[ -z "$MODEL" ]] && die "hyperopt preset requires --model"
    validate_model "$MODEL"
    ;;
  *)
    die "Unknown preset '${PRESET}'. Valid: quick-test, single-model, full-ensemble, hyperopt"
    ;;
esac

# --- Build training arguments for a given model ----------------------------
build_args() {
  local model="$1"
  local binary="${MODEL_BINARY[$model]}"
  local args=("$binary")

  args+=("--symbol" "$SYMBOL")
  args+=("--max-steps-per-epoch" "$MAX_STEPS")

  if [[ "$PRESET" == "hyperopt" ]]; then
    args+=("--trials" "$TRIALS")
  fi

  echo "${args[*]}"
}

# --- Generate K8s Job manifest --------------------------------------------
generate_job_manifest() {
  local model="$1"
  local job_name="train-${model}-${TIMESTAMP}"
  local train_args
  train_args="$(build_args "$model")"

  cat <<EOF
apiVersion: batch/v1
kind: Job
metadata:
  name: ${job_name}
  namespace: ${NAMESPACE}
  labels:
    foxhunt/job-type: training
    foxhunt/model: ${model}
    foxhunt/preset: ${PRESET}
spec:
  activeDeadlineSeconds: ${TIMEOUT}
  backoffLimit: 0
  template:
    metadata:
      labels:
        foxhunt/job-type: training
        foxhunt/model: ${model}
        foxhunt/preset: ${PRESET}
    spec:
      restartPolicy: Never
      nodeSelector:
        k8s.scaleway.com/pool-name: gpu-training
      tolerations:
        - key: nvidia.com/gpu
          operator: Exists
          effect: NoSchedule
      imagePullSecrets:
        - name: scw-registry
      containers:
        - name: training
          image: ${IMAGE}
          command: ["/bin/sh", "-c"]
          args:
            - "${train_args}"
          resources:
            requests:
              nvidia.com/gpu: "1"
              cpu: "4"
              memory: "16Gi"
            limits:
              nvidia.com/gpu: "1"
              cpu: "8"
              memory: "32Gi"
          volumeMounts:
            - name: training-data
              mountPath: /data
              readOnly: true
            - name: output
              mountPath: /output
      volumes:
        - name: training-data
          persistentVolumeClaim:
            claimName: training-data-pvc
        - name: output
          emptyDir: {}
EOF
}

# --- Submit a single job and collect its name ------------------------------
submitted_jobs=()

submit_job() {
  local model="$1"
  local job_name="train-${model}-${TIMESTAMP}"

  echo "--- Submitting job: ${job_name} (model=${model}, preset=${PRESET})"
  generate_job_manifest "$model" | kubectl apply -f -
  submitted_jobs+=("$job_name")
}

# --- Main dispatch ---------------------------------------------------------
echo "======================================================================"
echo "  Foxhunt Training Orchestrator"
echo "  Preset : ${PRESET}"
echo "  Symbol : ${SYMBOL}"
echo "  Image  : ${IMAGE}"
echo "  NS     : ${NAMESPACE}"
echo "======================================================================"
echo ""

case "$PRESET" in
  quick-test|single-model|hyperopt)
    submit_job "$MODEL"
    ;;
  full-ensemble)
    for m in "${ALL_MODELS[@]}"; do
      submit_job "$m"
      echo "  Waiting for ${m} to complete before next model..."
      kubectl -n "${NAMESPACE}" wait --for=condition=complete "job/train-${m}-${TIMESTAMP}" --timeout="${TIMEOUT}s" 2>/dev/null || true
    done
    ;;
esac

echo ""
echo "======================================================================"
echo "  All jobs submitted."
echo "======================================================================"
echo ""
echo "Monitor with:"
echo ""
for jn in "${submitted_jobs[@]}"; do
  echo "  kubectl -n ${NAMESPACE} get job ${jn}"
  echo "  kubectl -n ${NAMESPACE} logs job/${jn} -f"
  echo ""
done
echo "  kubectl -n ${NAMESPACE} get jobs -l foxhunt/preset=${PRESET}"
echo "  kubectl -n ${NAMESPACE} get pods -l foxhunt/job-type=training --watch"