introducing possibility to use identity for metamodel (usefull for metabolic...

introducing possibility to use identity for metamodel (usefull for metabolic seeds for which FBA always predict for influx the same value than the constraint applied for it)

src/dfba_sampling/metamodel.py

@@ -9,6 +9,7 @@ import itertools as it
 
 import numpy as np
 import pandas as pd
+import yaml
 from scipy.linalg import sqrtm
 from sklearn.metrics.pairwise import pairwise_kernels
 from sklearn.preprocessing import MinMaxScaler
@@ -78,6 +79,7 @@ class Metamodel(object):
         kernel_type="Matern",
         index_col=None,
         exp=None,
+        metamodel_config=None,
     ):
         self.kernel_type = kernel_type
         self.bact = bact
@@ -96,6 +98,11 @@ class Metamodel(object):
             self.Nobs, self.Nsub = np.shape(self.C_learn)
         if fluxes is not None:
             self.new_fluxlearningset(fluxes, index_col=index_col)
+        if metamodel_config is not None:
+            with open(metamodel_config, "r") as file:  # Load the user parameter yml
+                self.metamodel_yml = yaml.safe_load(file)
+        else:
+            self.metamodel_yml = None
 
     def new_constraintlearningset(
         self,
@@ -450,24 +457,33 @@ class Metamodel(object):
                 kernel_singleton_OrderOne[:, :, s] = self.ka0_predict(c, s)
             output = np.zeros([Npred, self.Ncompound])
             for comp in range(self.Ncompound):
-                kernel_singleton = np.zeros([Npred, self.Nobs, len(self.kept_P[comp])])
-                for a in range(len(self.kept_P[comp])):
-                    if len(self.kept_P[comp][a]) == 1:
-                        kernel_singleton[:, :, a] = kernel_singleton_OrderOne[
-                            :, :, self.kept_P[comp][a][0]
-                        ]
-                    else:
-                        kernel_singleton[:, :, a] = np.prod(
-                            kernel_singleton_OrderOne[:, :, self.kept_P[comp][a]],
-                            axis=2,
-                        )
-                output[:, comp] = self.theta_0[comp] + np.sum(
-                    [
-                        kernel_singleton[:, :, iv].dot(self.theta[comp][:, iv])
-                        for iv in range(len(self.kept_P[comp]))
-                    ],
-                    axis=0,
-                )
+                if (
+                    self.metamodel_yml is not None
+                    and "identity" in self.metamodel_yml[self.bact]
+                    and comp in self.metamodel_yml[self.bact]["identity"]
+                ):
+                    output[:, comp] = c[:, comp]
+                else:
+                    kernel_singleton = np.zeros(
+                        [Npred, self.Nobs, len(self.kept_P[comp])]
+                    )
+                    for a in range(len(self.kept_P[comp])):
+                        if len(self.kept_P[comp][a]) == 1:
+                            kernel_singleton[:, :, a] = kernel_singleton_OrderOne[
+                                :, :, self.kept_P[comp][a][0]
+                            ]
+                        else:
+                            kernel_singleton[:, :, a] = np.prod(
+                                kernel_singleton_OrderOne[:, :, self.kept_P[comp][a]],
+                                axis=2,
+                            )
+                    output[:, comp] = self.theta_0[comp] + np.sum(
+                        [
+                            kernel_singleton[:, :, iv].dot(self.theta[comp][:, iv])
+                            for iv in range(len(self.kept_P[comp]))
+                        ],
+                        axis=0,
+                    )
             if self.Scaler_fluxes_flag != "None":
                 output = self.Scaler_fluxes.inverse_transform(output)