From 8a5203accca1fde5b69c8b15ce7eb074d61e0919 Mon Sep 17 00:00:00 2001
From: Ysobel <ysobelsims@gmail.com>
Date: Thu, 22 Jan 2026 13:27:52 +1100
Subject: [PATCH] model path, train or no train

---
 README.md        | 23 +++++++++++++++++++++++
 ecosci/models.py | 15 +++++++++++++--
 2 files changed, 36 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index f36d463..53ba3f4 100644
--- a/README.md
+++ b/README.md
@@ -25,7 +25,9 @@ EcoNetToolkit lets you train a shallow neural network or classical models on you
     - [Inspecting Saved Models](#inspecting-saved-models)
   - [Config reference (YAML)](#config-reference-yaml)
     - [Simple example (single model, classification)](#simple-example-single-model-classification)
+    - [Multi-output (multi-target) prediction](#multi-output-multi-target-prediction)
     - [Available models and key parameters](#available-models-and-key-parameters)
+  - [Loading a model from a saved path](#loading-a-model-from-a-saved-path)
     - [Notes on metrics](#notes-on-metrics)
     - [Additional notes](#additional-notes)
   - [Hyperparameter Tuning](#hyperparameter-tuning)
@@ -306,6 +308,27 @@ See `configs/penguins_multilabel.yaml` and `configs/possum_multilabel.yaml` for
 **Linear Regression** (regression only)
 - `fit_intercept`: Whether to calculate the intercept (default: `true`)
 
+## Loading a model from a saved path
+
+You can load a previously trained model directly from a file by specifying `model_path` in the model's `params` section of your YAML config. This is useful for reusing or updating models without retraining.
+
+**Example:**
+
+```yaml
+models:
+  - name: random_forest
+    params:
+      model_path: outputs/possum/random_forest/model_random_forest_seed42.joblib
+      no_train: true  # If true, use the loaded model with no further training
+```
+
+- If `no_train: true`, EcoNetToolkit will use the loaded model for prediction only and will not retrain it with the current data.
+- If `no_train` is omitted or set to `false`, the loaded model will be further trained (fit) on the current data, allowing you to continue training or fine-tune.
+
+If `model_path` is provided, EcoNetToolkit will load the model from disk using joblib and use it for predictions or further evaluation. All other parameters are ignored when loading from a path and `no_train: true`.
+
+---
+
 ### Notes on metrics
 
 **Classification:**
diff --git a/ecosci/models.py b/ecosci/models.py
index fcf7491..24a3c8d 100644
--- a/ecosci/models.py
+++ b/ecosci/models.py
@@ -20,8 +20,8 @@
 class ModelZoo:
     @staticmethod
     def get_model(
-        name: str, 
-        problem_type: str = "classification", 
+        name: str,
+        problem_type: str = "classification",
         params: Dict[str, Any] = None,
         n_outputs: int = 1
     ):
@@ -39,6 +39,17 @@ def get_model(
             Number of output targets. If > 1, wraps the model in MultiOutput wrapper.
         """
         params = params or {}
+        # If model_path is provided, load the model from disk
+        model_path = params.get("model_path")
+        no_train = params.get("no_train", False)
+        if model_path:
+            import joblib
+            model = joblib.load(model_path)
+            # If no_train is True, return the loaded model directly
+            if no_train:
+                return model
+            # Otherwise, return the loaded model for further training (fit will be called)
+            return model
 
         if name.lower() == "mlp":
             from sklearn.neural_network import MLPClassifier, MLPRegressor