aug by sites

bin/train.sh

@@ -4,7 +4,7 @@ export CUDA_VISIBLE_DEVICES=2,3
 RUN_CONFIG=config.yml
 
 
-LOGDIR=/raid/bac/kaggle/logs/recursion_cell/test/c123_s12_smooth_nadam_/se_resnext50_32x4d/
+LOGDIR=/raid/bac/kaggle/logs/recursion_cell/test/c123_s1_smooth_nadam_rndsite/se_resnext50_32x4d/
 catalyst-dl run \
     --config=./configs/${RUN_CONFIG} \
     --logdir=$LOGDIR \

configs/config.yml

@@ -1,7 +1,7 @@
 model_params:
   model: cell_senet
   model_name: se_resnext50_32x4d
-  n_channels: 6
+  n_channels: 3
   num_classes: 1108
 
 args:
@@ -43,7 +43,7 @@ stages:
     train_csv: "./csv/train_0.csv"
     valid_csv: "./csv/valid_0.csv"
     root: "/raid/data/kaggle/recursion-cellular-image-classification/"
-    sites: [1, 2]
+    sites: [1]
     channels: [1, 2, 3]
 
   stage1:

src/dataset.py

@@ -225,7 +225,7 @@ def __init__(self,
         self.plates = df['plate'].values
         self.wells = df['well'].values
 
-        if mode == 'train':
+        if mode != 'test':
             self.labels = df['sirna'].values
         else:
             self.labels = [0] * len(self.experiments)
@@ -243,7 +243,15 @@ def __getitem__(self, idx):
 
         channel_paths = []
 
-        for site in self.sites:
+        if self.mode == 'train':
+            if np.random.rand() < 0.5:
+                sites = [1]
+            else:
+                sites = [2]
+        else:
+            sites = self.sites
+
+        for site in sites:
             for channel in self.channels:
                 path = image_path(
                     dataset=self.mode,
@@ -259,7 +267,7 @@ def __getitem__(self, idx):
         std_arr = []
         mean_arr = []
 
-        for site in self.sites:
+        for site in sites:
             for channel in self.channels:
                 mean = self.stat_dict[experiment][plate][well][site][channel]["mean"]
                 std = self.stat_dict[experiment][plate][well][site][channel]["std"]

src/make_submission.py

@@ -36,12 +36,12 @@ def predict_all():
     test_csv = '/raid/data/kaggle/recursion-cellular-image-classification/test.csv'
     # test_csv = './csv/valid_0.csv'
 
-    experiment = 'c123_s12_smooth_nadam_'
+    experiment = 'c123_s1_smooth_nadam_rndsite'
     model_name = 'se_resnext50_32x4d'
 
     log_dir = f"/raid/bac/kaggle/logs/recursion_cell/test/{experiment}/{model_name}/"
     root = "/raid/data/kaggle/recursion-cellular-image-classification/"
-    sites = [1, 2]
+    sites = [1]
     channels = [1,2,3]
 
     model = cell_senet(
@@ -55,26 +55,31 @@ def predict_all():
     model.load_state_dict(checkpoint['model_state_dict'])
     model = model.to(device)
 
-    # Dataset
-    dataset = RecursionCellularSite(
-        csv_file=test_csv,
-        root=root,
-        transform=valid_aug(512),
-        mode='test',
-        sites=sites,
-        channels=channels
-    )
-
-    loader = DataLoader(
-        dataset=dataset,
-        batch_size=128,
-        shuffle=False,
-        num_workers=4,
-    )
-
-    pred = predict(model, loader)
+    preds = []
 
-    all_preds = np.argmax(pred, axis=1)
+    for site in [1, 2]:
+        # Dataset
+        dataset = RecursionCellularSite(
+            csv_file=test_csv,
+            root=root,
+            transform=valid_aug(512),
+            mode='test',
+            sites=[site],
+            channels=channels
+        )
+
+        loader = DataLoader(
+            dataset=dataset,
+            batch_size=128,
+            shuffle=False,
+            num_workers=4,
+        )
+
+        pred = predict(model, loader)
+        preds.append(pred)
+
+    preds = np.asarray(preds).mean(axis=0)
+    all_preds = np.argmax(preds, axis=1)
     df = pd.read_csv(test_csv)
     submission = df.copy()
     submission['sirna'] = all_preds.astype(int)