deploy: braindecode/braindecode@6321baa

Author: bruAristimunha

master/_downloads/07fcc19ba03226cd3d83d4e40ec44385/auto_examples_python.zip

@@ -224,7 +224,7 @@
       },
       "outputs": [],
       "source": [
-        "from skorch.callbacks import LRScheduler\nfrom skorch.helper import predefined_split\n\nfrom braindecode.training import TimeSeriesLoss\nfrom braindecode import EEGRegressor\nfrom braindecode.training import CroppedTimeSeriesEpochScoring\n\n# These values we found good for shallow network for EEG MI decoding:\nlr = 0.0625 * 0.01\nweight_decay = 0\nbatch_size = 64\nn_epochs = 8\n\nregressor = EEGRegressor(\n    model,\n    cropped=True,\n    aggregate_predictions=False,\n    criterion=TimeSeriesLoss,\n    criterion__loss_function=torch.nn.functional.mse_loss,\n    optimizer=torch.optim.AdamW,\n    train_split=predefined_split(valid_set),\n    optimizer__lr=lr,\n    optimizer__weight_decay=weight_decay,\n    iterator_train__shuffle=True,\n    batch_size=batch_size,\n    callbacks=[\n        (\"lr_scheduler\", LRScheduler('CosineAnnealingLR', T_max=n_epochs - 1)),\n        ('r2_train', CroppedTimeSeriesEpochScoring(sklearn.metrics.r2_score,\n                                                   lower_is_better=False,\n                                                   on_train=True,\n                                                   name='r2_train')\n         ),\n        ('r2_valid', CroppedTimeSeriesEpochScoring(sklearn.metrics.r2_score,\n                                                   lower_is_better=False,\n                                                   on_train=False,\n                                                   name='r2_valid')\n         )\n    ],\n    device=device,\n)\nset_log_level(verbose='WARNING')"
+        "from skorch.callbacks import LRScheduler\nfrom skorch.helper import predefined_split\n\nfrom braindecode.training import TimeSeriesLoss\nfrom braindecode import EEGRegressor\nfrom braindecode.training import CroppedTimeSeriesEpochScoring\n\n# These values we found good for shallow network for EEG MI decoding:\nlr = 0.0625 * 0.01\nweight_decay = 0\nbatch_size = 27  # only 27 examples in train set, otherwise set to 64\nn_epochs = 8\n\nregressor = EEGRegressor(\n    model,\n    cropped=True,\n    aggregate_predictions=False,\n    criterion=TimeSeriesLoss,\n    criterion__loss_function=torch.nn.functional.mse_loss,\n    optimizer=torch.optim.AdamW,\n    train_split=predefined_split(valid_set),\n    optimizer__lr=lr,\n    optimizer__weight_decay=weight_decay,\n    iterator_train__shuffle=True,\n    batch_size=batch_size,\n    callbacks=[\n        (\"lr_scheduler\", LRScheduler('CosineAnnealingLR', T_max=n_epochs - 1)),\n        ('r2_train', CroppedTimeSeriesEpochScoring(sklearn.metrics.r2_score,\n                                                   lower_is_better=False,\n                                                   on_train=True,\n                                                   name='r2_train')\n         ),\n        ('r2_valid', CroppedTimeSeriesEpochScoring(sklearn.metrics.r2_score,\n                                                   lower_is_better=False,\n                                                   on_train=False,\n                                                   name='r2_valid')\n         )\n    ],\n    device=device,\n)\nset_log_level(verbose='WARNING')"
       ]
     },
     {

master/_downloads/22c027ac1764bd1e50c7597f407c49bb/plot_bcic_iv_4_ecog_cropped.ipynb

@@ -170,7 +170,7 @@
       },
       "outputs": [],
       "source": [
-        "import os\n\nfrom skorch.helper import predefined_split\nfrom skorch.callbacks import Checkpoint, EarlyStopping, EpochScoring\nfrom braindecode import EEGClassifier\n\nlr = 5e-3\nbatch_size = 512\nn_epochs = 25\nnum_workers = 0 if n_jobs <= 1 else n_jobs\n\ncp = Checkpoint(dirname='', f_criterion=None, f_optimizer=None, f_history=None)\nearly_stopping = EarlyStopping(patience=10)\ntrain_acc = EpochScoring(\n    scoring='accuracy', on_train=True, name='train_acc', lower_is_better=False)\nvalid_acc = EpochScoring(\n    scoring='accuracy', on_train=False, name='valid_acc',\n    lower_is_better=False)\ncallbacks = [\n    ('cp', cp),\n    ('patience', early_stopping),\n    ('train_acc', train_acc),\n    ('valid_acc', valid_acc)\n]\n\nclf = EEGClassifier(\n    model,\n    criterion=torch.nn.BCEWithLogitsLoss,\n    optimizer=torch.optim.Adam,\n    max_epochs=n_epochs,\n    iterator_train__shuffle=False,\n    iterator_train__sampler=train_sampler,\n    iterator_valid__sampler=valid_sampler,\n    iterator_train__num_workers=num_workers,\n    iterator_valid__num_workers=num_workers,\n    train_split=predefined_split(splitted['valid']),\n    optimizer__lr=lr,\n    batch_size=batch_size,\n    callbacks=callbacks,\n    device=device\n)\n# Model training for a specified number of epochs. `y` is None as it is already\n# supplied in the dataset.\nclf.fit(splitted['train'], y=None)\nclf.load_params(checkpoint=cp)  # Load the model with the lowest valid_loss\n\nos.remove('./params.pt')  # Delete parameters file"
+        "import os\n\nfrom skorch.helper import predefined_split\nfrom skorch.callbacks import Checkpoint, EarlyStopping, EpochScoring\nfrom braindecode import EEGClassifier\n\nlr = 5e-3\nbatch_size = 128  # 512 if data large enough\nn_epochs = 25\nnum_workers = 0 if n_jobs <= 1 else n_jobs\n\ncp = Checkpoint(dirname='', f_criterion=None, f_optimizer=None, f_history=None)\nearly_stopping = EarlyStopping(patience=10)\ntrain_acc = EpochScoring(\n    scoring='accuracy', on_train=True, name='train_acc', lower_is_better=False)\nvalid_acc = EpochScoring(\n    scoring='accuracy', on_train=False, name='valid_acc',\n    lower_is_better=False)\ncallbacks = [\n    ('cp', cp),\n    ('patience', early_stopping),\n    ('train_acc', train_acc),\n    ('valid_acc', valid_acc)\n]\n\nclf = EEGClassifier(\n    model,\n    criterion=torch.nn.BCEWithLogitsLoss,\n    optimizer=torch.optim.Adam,\n    max_epochs=n_epochs,\n    iterator_train__shuffle=False,\n    iterator_train__sampler=train_sampler,\n    iterator_valid__sampler=valid_sampler,\n    iterator_train__num_workers=num_workers,\n    iterator_valid__num_workers=num_workers,\n    train_split=predefined_split(splitted['valid']),\n    optimizer__lr=lr,\n    batch_size=batch_size,\n    callbacks=callbacks,\n    device=device\n)\n# Model training for a specified number of epochs. `y` is None as it is already\n# supplied in the dataset.\nclf.fit(splitted['train'], y=None)\nclf.load_params(checkpoint=cp)  # Load the model with the lowest valid_loss\n\nos.remove('./params.pt')  # Delete parameters file"
       ]
     },
     {

master/_downloads/3e0455c88adcaa2571230abb981020db/plot_relative_positioning.ipynb

@@ -287,7 +287,7 @@
 # These values we found good for shallow network for EEG MI decoding:
 lr = 0.0625 * 0.01
 weight_decay = 0
-batch_size = 64
+batch_size = 27  # only 27 examples in train set, otherwise set to 64
 n_epochs = 8
 
 regressor = EEGRegressor(

master/_downloads/6f1e7a639e0699d6164445b55e6c116d/auto_examples_jupyter.zip

@@ -307,7 +307,7 @@ def forward(self, x):
 from braindecode import EEGClassifier
 
 lr = 5e-3
-batch_size = 512
+batch_size = 128  # 512 if data large enough
 n_epochs = 25
 num_workers = 0 if n_jobs <= 1 else n_jobs