fullreplace pour l'automatisation des perturbations et les statistiques pour les composants

mechir/perturbation/GenericAutomatizedPerturbation.py

@@ -18,6 +18,8 @@
 # * `mot+` : ajout de `mot` au début du document ;
 # * `+mot` : ajout de `mot` à la fin du document.
 
+# La perturbation spéciale `full_replace` correspond à l'application de toutes les perturbations `replace` fournies appliquées en une fois.
+
 # Exemple :
 
 # ```
@@ -45,6 +47,7 @@
 
 # * `dot_model_name` : nom du modèle dense utilisé pour le calcul des représentations dans la base HuggingFace;
 # * `perturbations_path` : chemin vers le fichier de perturbations ;
+# * `full_replace` : indique si les perturbations fournies doivent être appliquées comme une seule perturbation `full_replace`ou non (/!\ Dans ce cas, les perturbations décrites dans `perturbations_path` doivent être toutes de type replace.)
 # * `data_path` : chemin vers le fichier CSV ou identifiant de la collection dans la librairie `ir_datasets`;
 # * `is_irdata` : indique si les données proviennent d'`ir_datasets` ;
 
@@ -96,15 +99,16 @@ text_field = "text"             # default : "text"
 is_irdata = False
 query_id_subset=["1"] # default : None
 
-perturbations_path = "perturbations_mainV2_elephantrose.txt"
+perturbations_path = "res/mE5-MainV2/exp6/test/perturbation_polysemie_mainV2.txt"
+full_replace = True
 gen_matrices = True
 gen_mean = True
 gen_std = True
-res_path = "res/mE5-MainV2/elephants-roses/" 
+res_path = "res/mE5-MainV2/exp6/test/" 
 
-load_values = True # default = False
-load_path = "res/mE5-MainV2/elephants-roses/perts_values.json"
-save_values = False # default = False
+load_values = False # default = False
+load_path = "res/mE5-MainV2/exp6/test/perts_values.json"
+save_values = True # default = False
 
 
 from mechir import Dot
@@ -136,6 +140,14 @@ def param_prepend(doc, mot="microwave"):
 def param_replace(doc, mot_orig="microwave", mot_rempl="toaster"):
     return doc.replace(mot_orig, mot_rempl)
 
+def param_full_replace(doc, replacmts):
+    res = doc
+    for replacemt in replacmts:
+        assert len(replacemt) == 3 , "Les perturbations de type full-replace doivent se composer de perturbations de type replace seulement"
+        res.replace(replacemt[1], replacemt[2])
+    return res
+
+
 
 # génération des perturbations à produire à partir du document
 def generer_transformations(fichier_regles):
@@ -464,12 +476,13 @@ def calculate_components(param_pert_dot_dataloader, gen_mean, gen_std):
         patching_head_outputs.append(patch_head_out)
     print(f"Nombre de NaN trouvés : {nan_cpt}")
     mean_head_outputs, std_head_outputs = None, None
+    patching_head_outputs_cleaned = torch.stack([tens.nan_to_num() for tens,_ in patching_head_outputs])
     if gen_mean :
-        mean_head_outputs = torch.mean(torch.stack([tens.nan_to_num() for tens,_ in patching_head_outputs]), axis=0).detach().to("cpu").numpy()
+        mean_head_outputs = torch.mean(patching_head_outputs_cleaned, axis=0).detach().to("cpu").numpy()
     if gen_std :
-        std_head_outputs = torch.std(torch.stack([tens.nan_to_num() for tens,_ in patching_head_outputs]), axis=0).detach().to("cpu").numpy()
+        std_head_outputs = torch.std(patching_head_outputs_cleaned, axis=0).detach().to("cpu").numpy()
 
-    return mean_head_outputs, std_head_outputs
+    return mean_head_outputs, std_head_outputs, patching_head_outputs_cleaned
 
 def load_dataset(data_path, is_irdata, query_id_subset = None, query_field = "text", text_field ="text"):
     if is_irdata:
@@ -481,6 +494,51 @@ def load_dataset(data_path, is_irdata, query_id_subset = None, query_field = "te
         # print("Number of query,doc pairs in dataset:", len(dataset))
     return dataset
 
+def one_pert_traitement(
+        pert, dot_model, 
+        dataset, 
+        full_replace = False, 
+        gen_matrices = False, 
+        gen_mean=False, 
+        gen_std=False
+    ):
+    if full_replace:
+        pert_fun = perturbation(lambda texte : param_full_replace(texte, pert))
+        pert_type = "replace"
+    elif pert[0]=="replace":
+        pert_fun = perturbation(lambda texte : param_replace(texte, pert[1], pert[2]))
+        pert_type = pert[0]
+    elif pert[0]=="prepend":
+        pert_fun = perturbation(lambda texte : param_prepend(texte, pert[1]))
+        pert_type = pert[0]
+    elif pert[0]=="append":
+        pert_fun = perturbation(lambda texte : param_append(texte, pert[1]))
+        pert_type = pert[0]
+    else:
+        raise RuntimeError("Unreachable")
+    param_pert_dot_collator = DotDataCollator(dot_model.tokenizer, pert_fun,  perturb_type=pert_type)
+    param_pert_dot_dataloader = DataLoader(dataset, collate_fn=param_pert_dot_collator)
+
+    # param_pert_batch = next(iter(param_pert_dot_dataloader))
+    # print("PERTURBATION",i,":", pert_name)
+    # pretty_print_triplets(param_pert_batch, dot_model.tokenizer, num=2)
+
+    baseline_perf = []
+    perturbed_perf = []
+    calculate_performance(dot_model, param_pert_dot_dataloader, baseline_perf, perturbed_perf, diags=True)
+
+    if gen_matrices:
+        mean_head_outputs, std_head_outputs, _ = calculate_components(param_pert_dot_dataloader, gen_mean, gen_std)
+
+    # Sauvegarde des valeurs pour la  perturbation
+    pert_values = {"baseline_perf": baseline_perf, "perturbed_perf": perturbed_perf}
+    if gen_matrices and mean_head_outputs is not None :
+        pert_values["patching_mean"] = mean_head_outputs
+    if gen_matrices and std_head_outputs is not None:
+        pert_values["patching_std"] = std_head_outputs
+
+    return pert_values
+
 # Calcul ou extraction des valeurs
 if load_values:
     perts_values = load_pert_result(load_path)
@@ -489,51 +547,32 @@ else:
     # Récup du modèle
     dot_model = Dot(dot_model_name)
 
-
     # Recup dataset
     dataset = load_dataset(data_path, is_irdata, query_id_subset, query_field, text_field)
 
-
     perts, perts_name = generer_transformations(perturbations_path)
     perts_values = {}
-    for i in range(len(perts)):
-        print(perts[i], "="*50)
-        if perts[i][0]=="replace":
-            pert_fun = perturbation(lambda texte : param_replace(texte, perts[i][1], perts[i][2]))
-        elif perts[i][0]=="prepend":
-            pert_fun = perturbation(lambda texte : param_prepend(texte, perts[i][1]))
-        elif perts[i][0]=="append":
-            pert_fun = perturbation(lambda texte : param_append(texte, perts[i][1]))
-        else:
-            raise RuntimeError("Unreachable")
-        param_pert_dot_collator = DotDataCollator(dot_model.tokenizer, pert_fun,  perturb_type=perts[i][0])
-        param_pert_dot_dataloader = DataLoader(dataset, collate_fn=param_pert_dot_collator)
-
-        # param_pert_batch = next(iter(param_pert_dot_dataloader))
-        # print("PERTURBATION",i,":", perts_name[i])
-        # pretty_print_triplets(param_pert_batch, dot_model.tokenizer, num=2)
-
-        baseline_perf = []
-        perturbed_perf = []
-        calculate_performance(dot_model, param_pert_dot_dataloader, baseline_perf, perturbed_perf, diags=True)
-
-        # plot_scores(baseline_perf, perturbed_perf, perts_name[i], save_path=res_path+"PerturbationScore_"+perts_name[i]+".png")
-
-        if gen_matrices:
-            mean_head_outputs, std_head_outputs = calculate_components(param_pert_dot_dataloader, gen_mean, gen_std)
-            # if mean_head_outputs is not None:
-            #     plot_components_V2(mean_head_outputs, title="Components Patching Results (mean) for "+ perts_name[i], save_path=res_path+"ComponentPatching-mean_"+perts_name[i]+"V2.png")
-            # if std_head_outputs is not None:
-            #     plot_components_V2(std_head_outputs, title="Components Patching Results (std) for "+ perts_name[i], save_path=res_path+"ComponentPatching-std_"+perts_name[i]+"V2.png")
-
-
-        # Sauvegarde des valeurs pour la i^eme perturbation
-
-        perts_values[perts_name[i]] = {"baseline_perf": baseline_perf, "perturbed_perf": perturbed_perf}
-        if gen_matrices and mean_head_outputs is not None :
-            perts_values[perts_name[i]]["patching_mean"] = mean_head_outputs
-        if gen_matrices and std_head_outputs is not None:
-            perts_values[perts_name[i]]["patching_std"] = std_head_outputs
+    if full_replace:
+        perts_values["full_replace"] = one_pert_traitement(
+            perts,
+            dot_model, 
+            dataset,
+            full_replace=True,
+            gen_matrices= gen_matrices,
+            gen_mean= gen_mean,
+            gen_std = gen_std)
+        perts_values["full_replace"]["descr"] = perts_name
+    else:
+        for i in range(len(perts)):
+            print(perts[i], "="*50)
+            perts_values[perts_name[i]] = one_pert_traitement(
+                perts[i], 
+                dot_model, 
+                dataset,
+                full_replace=False,
+                gen_matrices= gen_matrices,
+                gen_mean= gen_mean,
+                gen_std = gen_std)
 
     if save_values:
          save_pertubs_values(perts_values, f"{res_path}/perts_values.json")

mechir/tools/compte_noeuds_sensible.py

@@ -93,6 +93,8 @@ def perturbation_type(pert):
         return "append"
     if pert[-1] =="+":
         return "prepend" 
+    if pert == "full_replace":
+        return "full_replace"
     raise RuntimeWarning("Unreachable")
 
 def stats(noeuds_perts, path):