Refactor dependencies and installation for modular environments

- Update `pyproject.toml`:
  - Move heavy ML libraries (TensorFlow, Torch, H2O, etc.) from base dependencies to a new `standard` extra.
  - Define `ts` extra for time-series specific libraries.
  - Pin `scikit-learn` to 1.6.x and `numpy` < 2.0.0.
- Update `install.sh`:
  - Default to installing `[test,standard]`.
  - Add pre-installation step for heavy dependencies to improve pip resolution speed.
  - Implement constraint-based installation for time-series mode.
- Add `pytest.ini`: Configure to exclude time-series tests by default.
- Add `run_ts_tests.sh`: Helper script for running time-series tests.

Author: SamoraHunter

install.sh

@@ -2,7 +2,7 @@
 
 # Default values for standard installation
 ENV_NAME="ml_grid_env"
-EXTRAS="[test]"
+EXTRAS="[test,standard]"
 INSTALL_TYPE="standard"
 PROXY_MODE=false
 
@@ -86,16 +86,39 @@ python -m pip install --upgrade pip 'setuptools<70.0.0' wheel "${pip_base_args[@
 # Increase the resolver backtrack depth to handle the large, complex dependency
 # graph (TensorFlow + PyTorch + AutoGluon + FastAI + aeon[all_extras] etc.).
 # PIP_RESOLVER_MAX_BACKTRACK is honoured by pip >= 23.3.
-export PIP_RESOLVER_MAX_BACKTRACK=10000
+export PIP_RESOLVER_MAX_BACKTRACK=1000000
 
 # Pre-install heavy dependencies to simplify graph resolution
 echo "Pre-installing heavy dependencies..."
-python -m pip install "tensorflow==2.20.0" "torch==2.10.0" "nvidia-cuda-nvcc-cu12" "h2o>=3.46.0.5" "scikit-learn>=1.5.2,<1.6" "${pip_base_args[@]}" || print_error_and_exit "Failed to pre-install frameworks."
+python -m pip install "tensorflow==2.20.0" "torch==2.10.0" "nvidia-cuda-nvcc-cu12" "h2o>=3.46.0.5" "scikit-learn>=1.6.0,<1.7" "${pip_base_args[@]}" || print_error_and_exit "Failed to pre-install frameworks."
+
+if [ "$INSTALL_TYPE" = "time-series" ]; then
+    echo "Pre-installing time-series dependencies to simplify graph resolution..."
+    CONSTRAINTS_FILE="ts-constraints.txt"
+
+    # Pre-install the top-level heavy time-series libraries.
+    # This allows pip to resolve their complex dependencies in a dedicated step
+    # before the final project installation. These are the main libraries
+    # from the '[project.optional-dependencies].ts' section in pyproject.toml.
+    python -m pip install \
+        "aeon>=1.2.0" \
+        "tsfresh" \
+        "prophet==1.1.3" \
+        "pmdarima==2.0.3" \
+        "gluonts>=0.14.0" \
+        "${pip_base_args[@]}" || print_error_and_exit "Failed to install heavy time-series libraries."
+
+    echo "Generating constraints file to pin pre-installed packages..."
+    pip freeze > "$CONSTRAINTS_FILE"
+fi
 
 # Install the project in editable mode along with testing dependencies.
 # This reads all dependencies from pyproject.toml.
 echo "Installing project dependencies ($EXTRAS)..."
 pip_install_args=("--no-build-isolation" "-e" ".$EXTRAS")
+if [ "$INSTALL_TYPE" = "time-series" ]; then
+    pip_install_args+=("-c" "$CONSTRAINTS_FILE")
+fi
 if [ "$PROXY_MODE" = true ]; then
     pip_install_args+=("--trusted-host" "dh-cap02" "-i" "http://dh-cap02:8008/mirrors/ml_binary_classification_gridsearch_hyperOpt" "--retries" "5" "--timeout" "60")
 fi

pyproject.toml

@@ -24,38 +24,17 @@ dependencies = [
     # Core data and ML libraries
     "pandas>=2.2.0,<2.4.0",
     "polars", #==1.9.0
-    "numpy>=1.26.4,<2.0.0",
-    "scikit-learn>=1.5.2,<1.6",
+    "numpy>=1.26.4,<2.0.0", # Pinned to avoid breaking changes in v2
+    "scikit-learn>=1.6.0,<1.7", # Pinned to support autogluon.timeseries dependencies
     "imbalanced-learn>=0.12.4",
-
-    # ML frameworks and dependencies
-    "keras>=3.0.0",
-    "tensorflow>=2.16.0", # Relaxed to allow resolution
-    "torch>=2.2.0",      # Relaxed to allow resolution
-    "torchvision>=0.15.0",  # Updated: compatible with torch 2.x
-    "tab_transformer_pytorch>=0.3.0",
-    "lightgbm>=4.5.0",
-    "xgboost>=2.1.1",
-    "catboost>=1.2.7",
-    "h2o>=3.46.0.5",
     "scikeras>=0.13.0",
-    "autogluon.tabular",
-    "tpot==0.12.2",
-    "FLAML",
-    "autokeras",
-    "spacy>=3.8.0", # Force modern spacy for pydantic v2 compatibility
-    "fastai",
-    "ray>=2.43.0,<2.53.0",
 
     # Hyperparameter optimization and feature tools
     "hyperopt",#==0.2.7",
     "scikit-optimize>=0.10.2",
     "shap>=0.46.0",
     "pyimpetus>=4.1.2",
-    "fitter>=1.7.1",
     "fuzzysearch>=0.7.3",
-    # "simbsig",  # Commented out: conflicts with torch 2.x requirement
-    "tabpfn",
 
     # Visualization and notebook integration
     "matplotlib>=3.9.0",
@@ -79,6 +58,8 @@ dependencies = [
     "black",
     "ruff",
 
+
+
 ]
 
 [project.urls]
@@ -90,13 +71,31 @@ test = [
     "pytest>=7.0.0",
     "nbmake==1.4.6",
 ]
+standard = [
+    # Standard (non-time-series) AutoML and advanced libraries
+    "autogluon.tabular",
+    "tpot==0.12.2",
+    "FLAML",
+    "autokeras",
+    "spacy>=3.8.0",
+    "fastai",
+    "tab_transformer_pytorch>=0.3.0",
+    "tabpfn",
+    "fitter>=1.7.1",
+    "keras>=3.0.0",
+    "tensorflow>=2.16.0", # Relaxed to allow resolution
+    "torch>=2.2.0",      # Relaxed to allow resolution
+    "torchvision>=0.15.0",
+    "lightgbm>=4.5.0",
+    "xgboost>=2.1.1",
+    "catboost>=1.2.7",
+    "h2o>=3.46.0.5",
+    "ray>=2.43.0,<2.53.0",
+]
 ts = [
     # Time-series specific libraries
-    "aeon[all_extras]>=1.2.0", # Updated for compatibility with pandas>=2.2
-    # Removed tensorflow-addons: deprecated and incompatible with Python 3.12
+    "aeon>=1.2.0", # Core aeon toolkit
     "tsfresh",
-    "tensorflow_probability",
-    "keras-self-attention==0.51.0",
     "prophet==1.1.3",
     "pmdarima==2.0.3",
     "gluonts>=0.14.0",

pytest.ini

@@ -0,0 +1,5 @@
+[pytest]
+markers =
+    time_series: marks tests as time-series related
+    slow: marks tests as slow to run
+addopts = -m "not time_series and not slow" --ignore=tests/test_time_series_models.py

run_ts_tests.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+# Script to run only the time series unit tests
+pytest tests/test_time_series_models.py -m time_series