Add Docker setup and build PyG extensions from source for ROCm

- Install build tools (build-essential, cmake, ninja-build) in Dockerfile
- Compile torch-scatter, torch-sparse, torch-cluster, torch-spline-conv
  from source since ROCm 5.6 lacks prebuilt wheels
- Add Docker Compose usage instructions to README with build, run,
  and service management commands

Dockerfile

@@ -28,6 +28,10 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     libxrender-dev \
     libgomp1 \
     libnuma1 \
+    # build tools required for compiling PyG extensions from source
+    build-essential \
+    cmake \
+    ninja-build \
     && rm -rf /var/lib/apt/lists/*
 
 # Make python3.10 the default python3
@@ -47,15 +51,24 @@ RUN pip install --no-cache-dir \
     --index-url https://download.pytorch.org/whl/rocm5.6
 
 # ---------------------------------------------------------------------------
-# Install PyTorch Geometric and extensions with ROCm 5.6 wheels
+# Install PyTorch Geometric and extensions
 # ---------------------------------------------------------------------------
+# ROCm 5.6 does not ship prebuilt PyG extension wheels, so we compile them
+# from source.  torch-geometric itself is pure-Python and installs from PyPI.
+# The extensions (scatter, sparse, cluster, spline-conv) are built with
+# torch.utils.cpp_extension which automatically picks up the ROCm toolchain
+# shipped in the base image.
+# ---------------------------------------------------------------------------
+ENV ROCM_HOME=/opt/rocm
+ENV HIP_HOME=/opt/rocm
+
+RUN pip install --no-cache-dir torch-geometric==2.4.0
+
 RUN pip install --no-cache-dir \
-    torch-geometric==2.4.0 \
-    torch-scatter==2.1.2+pt21rocm5.6 \
-    torch-sparse==0.6.18+pt21rocm5.6 \
-    torch-cluster==1.6.2+pt21rocm5.6 \
-    torch-spline-conv==1.2.2+pt21rocm5.6 \
-    -f https://data.pyg.org/whl/torch-2.1.0+rocm5.6.html
+    torch-scatter==2.1.2 \
+    torch-sparse==0.6.18 \
+    torch-cluster==1.6.2 \
+    torch-spline-conv==1.2.2
 
 # ---------------------------------------------------------------------------
 # Install remaining Python dependencies from PyPI

README.md

@@ -210,6 +210,50 @@ pip install -r requirements.txt
 # pip install torch-geometric torch-scatter torch-sparse -f https://data.pyg.org/whl/torch-2.1.0+rocm5.6.html
 ```
 
+### Docker (Recommended)
+
+A pre-configured Docker image with ROCm 5.6 and all Python dependencies is provided for a reproducible, containerized environment.
+
+#### Prerequisites
+
+- Docker 20.10+ with Compose v2 (`docker compose`)
+- AMD GPU with ROCm 5.6+ drivers installed on the host
+- `video` and `render` group membership for GPU device access
+
+#### Build & Run
+
+```bash
+# 1. Build the image
+docker compose build
+
+# 2. Start the web dashboard (http://localhost:8000)
+docker compose up -d dashboard
+
+# 3. Run model training (one-off)
+docker compose run --rm train
+
+# 4. Start the live trading engine
+docker compose --profile live up -d live-trading
+```
+
+#### Useful Commands
+
+```bash
+# View dashboard logs
+docker compose logs -f dashboard
+
+# Open an interactive shell inside the container
+docker compose run --rm train bash
+
+# Stop all services
+docker compose down
+
+# Stop and remove live-trading service
+docker compose --profile live down
+```
+
+> **Note:** The container mounts `./data`, `./models`, and `./logs` as volumes so that datasets, trained models, and log files persist across restarts.
+
 ### Run Training
 
 ```bash