Dub

.github/workflows/draft_pdf.yml

@@ -14,10 +14,10 @@ jobs:
           # This should be the path to the paper within your repo.
           paper-path: paper/paper.md
       - name: Upload
-        uses: actions/upload-artifact@v1
+        uses: actions/upload-artifact@v4
         with:
           name: paper
           # This is the output path where Pandoc will write the compiled
           # PDF. Note, this should be the same directory as the input
           # paper.md
-          path: paper/paper.pdf
+          path: paper/paper.pdf

pynm/models/gamlss.py

@@ -176,9 +176,10 @@ def fit(self,train_data):
                                 family={self.family},
                                 data=train_data,
                                 method={self.method})''')
-
-    def predict(self,test_data,what='mu'):
-        """Predict from fitted gamlss model.
+
+
+    def predict(self, test_data,score, what='mu'):
+        """Predict from fitted gamlss model and retrieve AIC and BIC.
 
         Parameters
         ----------
@@ -189,6 +190,45 @@ def predict(self,test_data,what='mu'):
         """
         ro.globalenv['model'] = self.model
         ro.globalenv['test_data'] = test_data
+        params = r(f'''predictAll(model,newdata=test_data)''')
+
+
+        ro.globalenv['params'] = params
+
+        percentiles = r(f'p{self.family}(test_data${score}, params$mu, params$sigma, params$nu, params$tau)')
+        ro.globalenv['percentiles'] = percentiles
+
+        mu_pred = r(f'''predict(model, newdata=test_data, parameter="{what}")''')
+        #sigma_pred = r(f'''predict(model, newdata=test_data, parameter="{what}", se.fit=TRUE)''')
+        sigma_pred= r(f'''predict(model, newdata=test_data, parameter="sigma", type="response")''')
+        res = r(f'''resid(model)''')
+
+
+        quantiles = r(f'q{self.family}(pnorm(1), params$mu,params$sigma, params$nu,params$tau)')
+        ro.globalenv['quantiles'] = quantiles
+
+        sigma = r(f'''params$sigma''')
+        tau = r(f'''params$tau''')
+        nu = r(f'''params$nu''')
+        z_scores = r(f'''qnorm(percentiles)''')
+
+        # Compute AIC and BIC
+        aic = r('AIC(model)')
+        bic = r('BIC(model)')
+        if self.family == 'GA':
+            q_lower = r(f'q{self.family}(0.025, params$mu, params$sigma)')
+            q_upper = r(f'q{self.family}(0.975, params$mu, params$sigma)')
+        else:
+            q_lower = r(f'q{self.family}(0.025, params$mu, params$sigma, params$nu, params$tau)')
+            q_upper = r(f'q{self.family}(0.975, params$mu, params$sigma, params$nu, params$tau)')
+        prediction_interval = ro.r['data.frame'](lower=q_lower, upper=q_upper)
+
+        return mu_pred,sigma_pred, res, z_scores, q_lower, q_upper, sigma, tau, nu, aic, bic
+
+    def summary(self):
+        """Generate and return the summary of the fitted GAMLSS model."""
+        ro.globalenv['fitted_model'] = self.model  # Store the fitted model in the R global environment
+        model_summary = r('summary(fitted_model)')  # Use R's summary function on the fitted model
 
-        res = r(f'''predict(model,newdata=test_data,parameter="{what}")''')
-        return res
+        print(model_summary)  # Print the summary to the console
+        return model_summary  # Return the summary object for further use or inspection

pynm/pynm.py

@@ -160,6 +160,9 @@ def __init__(self, data, score, group, confounds,
         self.RMSE_GAMLSS = None
         self.SMSE_GAMLSS = None
         self.MSLL_GAMLSS = None
+        self.AIC_GAMLSS = None
+        self.BIC_GAMLSS = None
+
 
         if seed is not None:
             np.random.seed(seed)
@@ -733,7 +736,9 @@ def _svgp_normative_model(self,conf_mat,score,ctr_mask,nu=2.5,length_scale=1,len
                 y_pred = means.numpy()
                 y_true = score
                 residuals = (y_true - y_pred).astype(float)
-
+
+            self.AIC_GAMLSS=aic 
+            self.BIC_GAMLSS=bic
             self.data['GP_pred'] = y_pred
             self.data['GP_sigma'] = sigma.numpy()
             self.data['GP_residuals'] = residuals
@@ -744,7 +749,7 @@ def _svgp_normative_model(self,conf_mat,score,ctr_mask,nu=2.5,length_scale=1,len
             self.MSLL_GP = msll
 
 
-    def gamlss_normative_model(self,mu=None,sigma=None,nu=None,tau=None,family='SHASHo2',method='RS',cv_folds=1):
+    def gamlss_normative_model(self,mu=None,sigma=None,nu=None,tau=None,train_col=None,family='SHASHo2',method='RS',cv_folds=1):
         """Compute GAMLSS normative model.
 
         Parameters
@@ -785,13 +790,13 @@ def gamlss_normative_model(self,mu=None,sigma=None,nu=None,tau=None,family='SHAS
             nan_cols = ['LOESS_pred','LOESS_residuals','LOESS_z','LOESS_rank','LOESS_sigma',
             'Centiles_pred','Centiles_residuals','Centiles_z','Centiles','Centiles_rank','Centiles_sigma',
             'Centiles_95','Centiles_5','Centiles_32','Centiles_68']
-            gamlss_data = self.data[[c for c in self.data.columns if c not in nan_cols]]
-
+            gamlss_data = self.data[[c for c in train_col if c not in nan_cols]]
+            
             if cv_folds == 1:
                 gamlss.fit(gamlss_data[ctr_mask])
 
-                mu_pred = gamlss.predict(gamlss_data,what='mu')
-                sigma_pred = gamlss.predict(gamlss_data,what='sigma')
+                mu_pred,sigma_pred, res,z_scores, q_lower, q_upper, sigma, tau, nu, aic, bic = gamlss.predict(gamlss_data,self.score,what='mu')
+
 
                 # For MSLL
                 y_train_mean = np.mean(self.data[self.score].values[ctr_mask])
@@ -818,8 +823,8 @@ def gamlss_normative_model(self,mu=None,sigma=None,nu=None,tau=None,family='SHAS
 
                     gamlss.fit(X_train)
 
-                    cv_mu_pred = gamlss.predict(X_test,what='mu')
-                    cv_sigma_pred = gamlss.predict(X_test,what='sigma')
+                    cv_mu_pred,cv_sigma_pred, res, z_scores, q_lower, q_upper, sigma, tau, nu, aic, bic = gamlss.predict(X_test)
+
 
                     r = RMSE(X_test[self.score].values,cv_mu_pred)
                     s = SMSE(X_test[self.score].values,cv_mu_pred)
@@ -840,14 +845,18 @@ def gamlss_normative_model(self,mu=None,sigma=None,nu=None,tau=None,family='SHAS
                 mu_pred = gamlss.predict(gamlss_data,what='mu')
                 sigma_pred = gamlss.predict(gamlss_data,what='sigma')
 
-            self.data['GAMLSS_pred'] = mu_pred
+            self.data['GAMLSS_pred'] = mu_pred 
             self.data['GAMLSS_sigma'] = sigma_pred
-            self.data['GAMLSS_residuals'] = self.data[self.score] - self.data['GAMLSS_pred']
-            self.data['GAMLSS_z'] = self.data['GAMLSS_residuals']/self.data['GAMLSS_sigma']
+            self.data['GAMLSS_z'] = z_scores
+            self.data['GAMLSS_lower_error'] = q_lower
+            self.data['GAMLSS_upper_error'] = q_upper
 
             self.RMSE_GAMLSS = rmse
             self.SMSE_GAMLSS = smse
             self.MSLL_GAMLSS = msll
+
+            self.AIC_GAMLSS = aic
+            self.BIC_GAMLSS = bic
 
     def report(self):
         """ Prints the values of each metric (SMSE, RMSE, MSLL) for the models that have been run.