Added manual threhold choce and update an asample

Author: leschultz

examples/jupyter/tutorial_1.ipynb

@@ -253,8 +253,14 @@
     "id": "HCL2WngJWphB"
    },
    "source": [
-    "# Assessing and Fitting the Model\n",
-    "Now, we assess the model through cross valication and then fit a final model on all data that can be used by a researcher. The assessment of the model is saved in a directory of the user's choice."
+    "# Assessing and Fitting the Model"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We can fit a single model without assessment, which is faster because of no nested cross validation. However, overfitting may occur."
    ]
   },
   {
@@ -266,6 +272,22 @@
    "outputs": [],
    "source": [
     "model = domain_model(gs_model, ds_model, uq_model, splits)\n",
+    "model.fit(X, y, g)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We can assess the model through cross validation and then fit a final model on all data that can be used by a researcher. The assessment of the model is saved in a directory of the user's choice. The fitting from the previous step is automatically performed after the nested cross validation to provide a usable model."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
     "cv = nested_cv(X, y, g, model, splits, save='./runs')\n",
     "_, model = cv.assess()"
    ]
@@ -291,6 +313,23 @@
     "df = model.predict(X)\n",
     "print(df)"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Maybe the predefined thresholds for domain are insufficient. We can instead use some manual thresholds as a list of douples with <('dissimilarity measure', 'domain of id or od', 'threshold')> as follows:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "df = model.predict(X, [('dist', 'id', 0.75), ('dist', 'id_bin', 0.2)])\n",
+    "print(df)"
+   ]
   }
  ],
  "metadata": {

src/madml/models/combine.py

@@ -44,7 +44,14 @@ def domain_pred(dist, dist_cut, domain):
     return do_pred
 
 
-def domain_preds(pred, dists, methods, thresholds, suffix=''):
+def domain_preds(
+                 pred,
+                 dists=None,
+                 methods=None,
+                 thresholds=None,
+                 suffix='',
+                 manual_th=None,
+                 ):
     '''
     The domain predictor based on thresholds.
 
@@ -59,27 +66,46 @@ def domain_preds(pred, dists, methods, thresholds, suffix=''):
         pred = The predictions of domain.
     '''
 
-    for i in dists:
+    if manual_th is not None:
+        for i in manual_th:
 
-        if suffix:
-            col = i+suffix
-        else:
-            col = i
+            dist, domain_type, th = i
+
+            if 'id' in domain_type:
+                domain = True
+            elif 'od' in domain_type:
+                domain = False
 
-        for j, k in zip([True, False], ['id', 'od']):
-            for method in methods:
-                k += method
-                for key, value in thresholds[i][k].items():
+            name = '{} by {} for {}'.format(domain_type.upper(), dist, th)
+            do_pred = domain_pred(
+                                  pred[dist],
+                                  th,
+                                  domain,
+                                  )
+
+            pred[name] = do_pred
+    else:
+        for i in dists:
 
-                    name = '{} by {} for {}'.format(k.upper(), i, key)
+            if suffix:
+                col = i+suffix
+            else:
+                col = i
 
-                    do_pred = domain_pred(
-                                          pred[col],
-                                          value['Threshold'],
-                                          j,
-                                          )
+            for j, k in zip([True, False], ['id', 'od']):
+                for method in methods:
+                    k += method
+                    for key, value in thresholds[i][k].items():
 
-                    pred[name] = do_pred
+                        name = '{} by {} for {}'.format(k.upper(), i, key)
+
+                        do_pred = domain_pred(
+                                              pred[col],
+                                              value['Threshold'],
+                                              j,
+                                              )
+
+                        pred[name] = do_pred
 
     return pred
 
@@ -359,12 +385,13 @@ def fit(self, X, y, g):
 
         return data_cv, data_cv_bin
 
-    def predict(self, X):
+    def predict(self, X, manual_th=None):
         '''
         Give domain classification along with other regression predictions.
 
         inputs:
             X = The features.
+            manual_th = Manual thresholds.
 
         outputs:
            pred = A pandas dataframe containing prediction data.
@@ -400,11 +427,19 @@ def predict(self, X):
             pred['y_stdc/std(y)'] = y_stdc_norm
 
         pred = pd.DataFrame(pred)
-        pred = domain_preds(
-                            pred,
-                            self.dists,
-                            self.methods,
-                            self.thresholds,
-                            )
+
+        if manual_th is None:
+            pred = domain_preds(
+                                pred,
+                                self.dists,
+                                self.methods,
+                                self.thresholds,
+                                )
+        else:
+
+            pred = domain_preds(
+                                pred,
+                                manual_th=manual_th
+                                )
 
         return pred