[DRAFT] Protocol redesign

benchmarks/MOABB/Commands_run_experiment.txt

@@ -0,0 +1 @@
+python run_experiments.py     --hparams=hparams/MotorImagery/BNCI2014001/EEGNet.yaml     --data_folder=./eeg_data     --output_folder=./results/test_run     --nsbj=9     --nsess=2     --seed=12346     --nruns=1     --train_mode=leave-one-session-out

benchmarks/MOABB/dataio/preprocessing.py

@@ -24,39 +24,82 @@ def to_tensor(epoch):
 cached_create_filter = cache(mne.filter.create_filter)
 
 
-@takes("epoch", "info", "target_sfreq", "fmin", "fmax")
-@provides("epoch", "sfreq", "target_sfreq", "fmin", "fmax")
-def bandpass_resample(epoch, info, target_sfreq, fmin, fmax):
-    """Bandpass filter and resample an epoch."""
-
-    bandpass = cached_create_filter(
-        None,
-        info["sfreq"],
-        l_freq=fmin,
-        h_freq=fmax,
-        method="fir",
-        fir_design="firwin",
-        verbose=False,
-    )
-
-    # Check that filter length is reasonable
-    filter_length = len(bandpass)
-    len_x = epoch.shape[-1]
-    if filter_length > len_x:
-        # TODO: These long filters result in massive performance degradation... Do we
-        #       want to throw an error instead? This usually happens when fmin is used
-        logging.warning(
-            "filter_length (%i) is longer than the signal (%i), "
-            "distortion is likely. Reduce filter length or filter a longer signal.",
-            filter_length,
-            len_x,
+def bandpass_resample():
+    @takes("epoch", "info", "target_sfreq", "fmin", "fmax")
+    @provides("epoch", "sfreq", "target_sfreq", "fmin", "fmax")
+    def _bandpass_resample(epoch, info, target_sfreq, fmin, fmax):
+        """Bandpass filter and resample an epoch."""
+
+        bandpass = cached_create_filter(
+            None,
+            info["sfreq"],
+            l_freq=fmin,
+            h_freq=fmax,
+            method="fir",
+            fir_design="firwin",
+            verbose=False,
+        )
+
+        # Check that filter length is reasonable
+        filter_length = len(bandpass)
+        len_x = epoch.shape[-1]
+        if filter_length > len_x:
+            # TODO: These long filters result in massive performance degradation... Do we
+            #       want to throw an error instead? This usually happens when fmin is used
+            logging.warning(
+                "filter_length (%i) is longer than the signal (%i), "
+                "distortion is likely. Reduce filter length or filter a longer signal.",
+                filter_length,
+                len_x,
+            )
+
+        yield mne.filter.resample(
+            epoch,
+            up=target_sfreq,
+            down=info["sfreq"],
+            method="polyphase",
+            window=bandpass,
+        )
+        yield target_sfreq
+
+    return _bandpass_resample
+
+
+'''def bandpass_resample(target_sfreq, fmin, fmax):
+    """Create a dynamic item that bandpass filters and resamples an epoch."""
+
+    @takes("epoch", "info")
+    @provides("epoch")
+    def _bandpass_resample(epoch, info):
+        bandpass = cached_create_filter(
+            None,
+            info["sfreq"],
+            l_freq=fmin,
+            h_freq=fmax,
+            method="fir",
+            fir_design="firwin",
+            verbose=False,
+        )
+        breakpoint()
+        # Check that filter length is reasonable
+        filter_length = len(bandpass)
+        len_x = epoch.shape[-1]
+        if filter_length > len_x:
+            logging.warning(
+                "filter_length (%i) is longer than the signal (%i), "
+                "distortion is likely. Reduce filter length or filter a longer signal.",
+                filter_length,
+                len_x,
+            )
+
+        filtered = mne.filter.resample(
+            epoch,
+            up=target_sfreq,
+            down=info["sfreq"],
+            method="polyphase",
+            window=bandpass,
         )
+        yield filtered
 
-    yield mne.filter.resample(
-        epoch,
-        up=target_sfreq,
-        down=info["sfreq"],
-        method="polyphase",
-        window=bandpass,
-    )
-    yield target_sfreq
+    return _bandpass_resample
+'''

benchmarks/MOABB/hparams/MotorImagery/BNCI2014001/EEGNet.yaml

@@ -1,6 +1,9 @@
 seed: 1234
 __set_torchseed: !apply:torch.manual_seed [!ref <seed>]
 
+#OVERRIDES
+
+
 # DIRECTORIES
 data_folder: !PLACEHOLDER  #'/path/to/dataset'. The dataset will be automatically downloaded in this folder
 cached_data_folder: !PLACEHOLDER #'path/to/pickled/dataset'
@@ -39,6 +42,34 @@ C: 22
 test_with: 'last' # 'last' or 'best'
 test_key: "acc" # Possible opts: "loss", "f1", "auc", "acc"
 
+# DATASET
+
+# Get target subject
+#target_subject: !ref <dataset>.subject_list[!ref <target_subject_idx>]#
+
+# Create the subjects list
+#subjects: [!ref <target_subject>]
+
+# Create dataset using EpochedEEGDataset
+#dataset_class: !new:dataio.datasets.EpochedEEGDataset
+
+#json_path: !apply:os.path.join [!ref <cached_data_folder>, "index.json"]
+#save_path: !ref <data_folder>
+#dynamic_items:
+#    - !name:dataio.preprocessing.to_tensor
+#output_keys: ["label", "subject", "session", "epoch"]
+#preload: True
+
+#from_moabb_datset: !apply: !ref <dataset_class>.from_moabb
+#    - !ref <dataset>
+#    - !ref <json_path>
+#    - !ref <subjects>
+#    - !ref <save_path>
+#    - !ref <dynamic_items>
+#    - !ref <output_keys>
+#    - !ref <preload>
+#    - !ref <tmin>
+#    - !ref <tmax>
 # METRICS
 f1: !name:sklearn.metrics.f1_score
     average: 'macro'
@@ -52,7 +83,7 @@ metrics:
 n_train_examples: 100  # it will be replaced in the train script
 # checkpoints to average
 avg_models: 10 # @orion_step1: --avg_models~"uniform(1, 15,discrete=True)"
-number_of_epochs: 862 # @orion_step1: --number_of_epochs~"uniform(250, 1000, discrete=True)"
+number_of_epochs: 800 # @orion_step1: --number_of_epochs~"uniform(250, 1000, discrete=True)"
 lr: 0.0001 # @orion_step1: --lr~"choices([0.01, 0.005, 0.001, 0.0005, 0.0001])"
 # Learning rate scheduling (cyclic learning rate is used here)
 max_lr: !ref <lr> # Upper bound of the cycle (max value of the lr)

benchmarks/MOABB/models/EEGNet.py

@@ -240,6 +240,7 @@ def forward(self, x):
         x : torch.Tensor (batch, time, EEG channel, channel)
             Input to convolve. 4d tensors are expected.
         """
+        x = x.transpose(1, 2)
         x = self.conv_module(x)
         x = self.dense_module(x)
         return x