set model_type in eval

eval.py

@@ -52,7 +52,8 @@ def parse_run_path(s):
         res["position"] = x[2]
         res["architecture"] = x[5]
         res["attention-scaling"] = x[6]
-        res["frozen"] = run_name.endswith("_frozen")
+        res["frozen"] = "_frozen" in run_name
+        res["wa"] = "_wa" in run_name
     return res
 
 
@@ -98,6 +99,10 @@ def eval_t1(run, e, dev_loader, test_loader):
     scal = run["attention-scaling"]
     frozen = run["frozen"]
     crps = run["corpus"]
+    if run["wa"]:
+        model_type = MiniBertForSequenceClassificationWithAttention
+    else:
+        model_type = MiniBertForSequenceClassification
 
     model, _, _, _ = t1_model_from_params(
         mlm_path(f"models/{crps}", d, att, pos, arch, scal),
@@ -112,7 +117,8 @@ def eval_t1(run, e, dev_loader, test_loader):
         checkpoint_path=str(Path(run["path"], f"checkpoint-{e:05}.tar")),
         height=height,
         depth=depth,
-        attention_scaling=parse_attention_scaling(scal)
+        attention_scaling=parse_attention_scaling(scal),
+        model_type=model_type
     )
     model.eval()
     dev_tp, dev_fp, dev_fn, dev_r, dev_p, dev_f = fmeasure_deft2018_t1(
@@ -149,6 +155,10 @@ def eval_t2(run, e, dev_loader, test_loader):
     scal = run["attention-scaling"]
     frozen = run["frozen"]
     crps = run["corpus"]
+    if run["wa"]:
+        model_type = MiniBertForSequenceClassificationWithAttention
+    else:
+        model_type = MiniBertForSequenceClassification
 
     model, _, _, _ = t2_model_from_params(
         mlm_path(f"models/{crps}", d, att, pos, arch, scal),
@@ -163,7 +173,8 @@ def eval_t2(run, e, dev_loader, test_loader):
         checkpoint_path=str(Path(run["path"], f"checkpoint-{e:05}.tar")),
         height=height,
         depth=depth,
-        attention_scaling=parse_attention_scaling(scal)
+        attention_scaling=parse_attention_scaling(scal),
+        model_type=model_type
     )
     model.eval()