add TS115 (#2)

* towards ts115 eval

* eval ts115

* fix api typo

Author: omarperacha

.gitattributes

@@ -1,3 +1,4 @@
 ps4_models/Conv/*.pt filter=lfs diff=lfs merge=lfs -text
 ps4_models/Mega/*.pt filter=lfs diff=lfs merge=lfs -text
 ps4_data/data/cb513/CB513_embeddings.npz filter=lfs diff=lfs merge=lfs -text
+ps4_data/data/ts115/TS115_embeddings.npz filter=lfs diff=lfs merge=lfs -text

main.py

@@ -54,15 +54,17 @@ def print_help():
                     if sys.argv[3] in ['--mega', '--conv']:
                         if sys.argv[3] == '--conv':
                             model_name = 'PS4_Conv'
-                            weights_path = 'ps4_models/Conv/PS4-Conv_epoch-5_loss-0.652_acc-77.905.pt'
+                            path = 'ps4_models/Conv/PS4-Conv_epoch-5_loss-0.652_acc-77.905.pt'
                     else:
                         print(f'Please specify a valid model name. found: {sys.argv[3]}')
 
                     if len(sys.argv) > 4:
                         path = sys.argv[4]
 
                 if sys.argv[2] == '--cb513':
-                    eval_cb513(path, model_name=model_name)
+                    eval_alt(path, ds_name='cb513', model_name=model_name)
+                elif sys.argv[2] == '--ts115':
+                    eval_alt(path, ds_name='ts115', model_name=model_name)
                 elif sys.argv[2] == '--ps4':
                     eval_ps4_test(path, model_name=model_name)
                 else:
@@ -103,4 +105,3 @@ def print_help():
 
 
 
-

ps4_data/data/ts115/TS115.fasta

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fcc6fa43810fb08438ac0d1987ec966ae318e84f3a5757ed4f60196670025d8
+size 112805135

ps4_data/data/ts115/TS115_HHblits.csv

@@ -250,10 +250,13 @@ def pt_2_csv(phase):
 
 
 # Mark: single sequence secondary structure
-def load_cb513_dataset():
+def load_alt_dataset(ds_name):
 
-    df = pd.read_csv(f'ps4_data/data/cb513/CB513_HHblits.csv')
-    embs = np.load(f'ps4_data/data/cb513/CB513_embeddings.npz', allow_pickle=True)
+    if ds_name.upper() not in ['CB513', 'TS115']:
+        raise ValueError(f"ds_name must be one of ['CB513', 'TS115'], got {ds_name}")
+
+    df = pd.read_csv(f'ps4_data/data/{ds_name.lower()}/{ds_name.upper()}_HHblits.csv')
+    embs = np.load(f'ps4_data/data/{ds_name.lower()}/{ds_name.upper()}_embeddings.npz', allow_pickle=True)
     for row in range(len(df)):
         res_string = df['input'][row]
 
@@ -265,7 +268,7 @@ def load_cb513_dataset():
         for c in mask_raw:
             mask_str += f'{c} '
 
-        _, y, mask = get_input_data_from_res_seq(res_string, ss_string, mask_str, 'CB513')
+        _, y, mask = get_input_data_from_res_seq(res_string, ss_string, mask_str)
         r = torch.from_numpy(embs[str(row)]).float()
         yield r, y, mask
 

ps4_data/data/ts115/TS115_embeddings.npz

@@ -50,7 +50,7 @@ def eval_ps4_test(load_path, model_name='PS4_Mega'):
           f"q3 acc: {sum(q3_accs)/len(whole_accs)}")
 
 
-def eval_cb513(load_path, use_mask=True, model_name='PS4_Mega'):
+def eval_alt(load_path, ds_name='cb513', use_mask=True, model_name='PS4_Mega'):
     print("starting")
     model = load_trained_model(load_path, model_name)
 
@@ -61,7 +61,7 @@ def eval_cb513(load_path, use_mask=True, model_name='PS4_Mega'):
     val_accs = []
     q3_accs = []
 
-    for r, y, mask in load_cb513_dataset():
+    for r, y, mask in load_alt_dataset(ds_name):
 
         count += 1
 
@@ -104,7 +104,7 @@ def eval_cb513(load_path, use_mask=True, model_name='PS4_Mega'):
             whole_accs.append(ss_acc)
             __q8_q3_from_confusion(ss_confusion)
 
-    print(f"\nDONE: CB513, whole q8 acc: {sum(whole_accs)/len(whole_accs)}\n"
+    print(f"\nDONE: {ds_name}, whole q8 acc: {sum(whole_accs)/len(whole_accs)}\n"
           f"val only q8 acc: {sum(val_accs)/len(val_accs)}\n"
           f"q3 acc: {sum(q3_accs)/len(val_accs)}")
 

ps4_data/utils.py

@@ -1,4 +1,4 @@
-from ps4_data.utils import load_data, load_cb513_dataset
+from ps4_data.utils import load_data, load_alt_dataset
 from torch import nn, cuda, load, device
 from ps4_models.classifiers import PS4_Mega, PS4_Conv
 
@@ -34,7 +34,8 @@ def __get_loader_for(ds):
     loader_dict = {
         'train': load_data('train'),
         'valid': load_data('valid'),
-        'cb513': load_cb513_dataset()
+        'cb513': load_alt_dataset('cb513'),
+        'ts115': load_alt_dataset('ts115')
     }
 
     return loader_dict[dataset]

ps4_eval/eval.py

@@ -8,5 +8,4 @@ Pillow~=8.1.2
 setuptools~=57.4.0
 pandas~=1.3.2
 wget~=3.2
-edlib~=1.3.8.post2
 maturin~=0.14.14