deploy: braindecode/braindecode@39c9bd5

Author: agramfort

master/_downloads/07fcc19ba03226cd3d83d4e40ec44385/auto_examples_python.zip

@@ -51,18 +51,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "## Create model and compute windowing parameters\n\n\n"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "collapsed": false
-      },
-      "outputs": [],
-      "source": [
-        "# In contrast to trialwise decoding, we first have to create the model\n# before we can cut the dataset into windows. This is because we need to\n# know the receptive field of the network to know how large the window\n# stride should be.\n#"
+        "## Create model and compute windowing parameters\n\nIn contrast to trialwise decoding, we first have to create the model\nbefore we can cut the dataset into windows. This is because we need to\nknow the receptive field of the network to know how large the window\nstride should be.\n\n\n"
       ]
     },
     {
@@ -134,7 +123,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "Cut the data into windows\n -------------------------\n#####################################################################\n In contrast to trialwise decoding, we have to supply an explicit window size and\n window stride to the ``create_windows_from_events`` function.\n\n\n"
+        "## Cut the data into windows\nIn contrast to trialwise decoding, we have to supply an explicit\nwindow size and window stride to the ``create_windows_from_events``\nfunction.\n\n\n"
       ]
     },
     {
@@ -188,7 +177,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "Plot Results\n ------------\n\n#####################################################################\n\n"
+        "## Plot Results\nThis is again the same code as in trialwise decoding.\n~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n<div class=\"alert alert-info\"><h4>Note</h4><p>Note that we drop further in the classification error and\n    loss as in the trialwise decoding tutorial.</p></div>\n\n\n"
       ]
     },
     {
@@ -199,7 +188,7 @@
       },
       "outputs": [],
       "source": [
-        "# This is again the same code as in trialwise decoding.\n#\n# .. note::\n#\n#     Note that we drop further in the classification error and\n#     loss as in the trialwise decoding tutorial.\n#\n\nimport matplotlib.pyplot as plt\nfrom matplotlib.lines import Line2D\nimport pandas as pd\n\n# Extract loss and accuracy values for plotting from history object\nresults_columns = ['train_loss', 'valid_loss', 'train_accuracy', 'valid_accuracy']\ndf = pd.DataFrame(clf.history[:, results_columns], columns=results_columns,\n                  index=clf.history[:, 'epoch'])\n\n# get percent of misclass for better visual comparison to loss\ndf = df.assign(train_misclass=100 - 100 * df.train_accuracy,\n               valid_misclass=100 - 100 * df.valid_accuracy)\n\nplt.style.use('seaborn')\nfig, ax1 = plt.subplots(figsize=(8, 3))\ndf.loc[:, ['train_loss', 'valid_loss']].plot(\n    ax=ax1, style=['-', ':'], marker='o', color='tab:blue', legend=False, fontsize=14)\n\nax1.tick_params(axis='y', labelcolor='tab:blue', labelsize=14)\nax1.set_ylabel(\"Loss\", color='tab:blue', fontsize=14)\n\nax2 = ax1.twinx()  # instantiate a second axes that shares the same x-axis\n\ndf.loc[:, ['train_misclass', 'valid_misclass']].plot(\n    ax=ax2, style=['-', ':'], marker='o', color='tab:red', legend=False)\nax2.tick_params(axis='y', labelcolor='tab:red', labelsize=14)\nax2.set_ylabel(\"Misclassification Rate [%]\", color='tab:red', fontsize=14)\nax2.set_ylim(ax2.get_ylim()[0], 85)  # make some room for legend\nax1.set_xlabel(\"Epoch\", fontsize=14)\n\n# where some data has already been plotted to ax\nhandles = []\nhandles.append(Line2D([0], [0], color='black', linewidth=1, linestyle='-', label='Train'))\nhandles.append(Line2D([0], [0], color='black', linewidth=1, linestyle=':', label='Valid'))\nplt.legend(handles, [h.get_label() for h in handles], fontsize=14)\nplt.tight_layout()"
+        "import matplotlib.pyplot as plt\nfrom matplotlib.lines import Line2D\nimport pandas as pd\n\n# Extract loss and accuracy values for plotting from history object\nresults_columns = ['train_loss', 'valid_loss', 'train_accuracy', 'valid_accuracy']\ndf = pd.DataFrame(clf.history[:, results_columns], columns=results_columns,\n                  index=clf.history[:, 'epoch'])\n\n# get percent of misclass for better visual comparison to loss\ndf = df.assign(train_misclass=100 - 100 * df.train_accuracy,\n               valid_misclass=100 - 100 * df.valid_accuracy)\n\nplt.style.use('seaborn')\nfig, ax1 = plt.subplots(figsize=(8, 3))\ndf.loc[:, ['train_loss', 'valid_loss']].plot(\n    ax=ax1, style=['-', ':'], marker='o', color='tab:blue', legend=False, fontsize=14)\n\nax1.tick_params(axis='y', labelcolor='tab:blue', labelsize=14)\nax1.set_ylabel(\"Loss\", color='tab:blue', fontsize=14)\n\nax2 = ax1.twinx()  # instantiate a second axes that shares the same x-axis\n\ndf.loc[:, ['train_misclass', 'valid_misclass']].plot(\n    ax=ax2, style=['-', ':'], marker='o', color='tab:red', legend=False)\nax2.tick_params(axis='y', labelcolor='tab:red', labelsize=14)\nax2.set_ylabel(\"Misclassification Rate [%]\", color='tab:red', fontsize=14)\nax2.set_ylim(ax2.get_ylim()[0], 85)  # make some room for legend\nax1.set_xlabel(\"Epoch\", fontsize=14)\n\n# where some data has already been plotted to ax\nhandles = []\nhandles.append(Line2D([0], [0], color='black', linewidth=1, linestyle='-', label='Train'))\nhandles.append(Line2D([0], [0], color='black', linewidth=1, linestyle=':', label='Valid'))\nplt.legend(handles, [h.get_label() for h in handles], fontsize=14)\nplt.tight_layout()"
       ]
     },
     {

master/_downloads/1224801107d6c3b138edf8ee33de0561/plot_bcic_iv_2a_moabb_cropped.ipynb

@@ -159,7 +159,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "## Training\n\nWe can now train our network on the pretext task. We use similar\nhyperparameters as in [1]_, but reduce the number of epochs and increase the\nlearning rate to account for the smaller setting of this example.\n\n"
+        "## Training\n\nWe can now train our network on the pretext task. We use similar\nhyperparameters as in [1]_, but reduce the number of epochs and\nincrease the learning rate to account for the smaller setting of\nthis example.\n\n"
       ]
     },
     {

master/_downloads/3e0455c88adcaa2571230abb981020db/plot_relative_positioning.ipynb

@@ -106,9 +106,6 @@
 # Create model and compute windowing parameters
 # ---------------------------------------------
 #
-
-######################################################################
-
 # In contrast to trialwise decoding, we first have to create the model
 # before we can cut the dataset into windows. This is because we need to
 # know the receptive field of the network to know how large the window
@@ -191,9 +188,9 @@
 ######################################################################
 # Cut the data into windows
 # -------------------------
-######################################################################
-# In contrast to trialwise decoding, we have to supply an explicit window size and
-# window stride to the ``create_windows_from_events`` function.
+# In contrast to trialwise decoding, we have to supply an explicit
+# window size and window stride to the ``create_windows_from_events``
+# function.
 #
 
 from braindecode.preprocessing import create_windows_from_events
@@ -289,12 +286,9 @@
 
 ######################################################################
 # Plot Results
-# ------------
-#
-######################################################################
-
+# ----------------
 # This is again the same code as in trialwise decoding.
-#
+# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 # .. note::
 #
 #     Note that we drop further in the classification error and

master/_downloads/6f1e7a639e0699d6164445b55e6c116d/auto_examples_jupyter.zip

@@ -4,7 +4,7 @@
 
 This tutorial shows how to search data augmentations using braindecode.
 Indeed, it is known that the best augmentation to use often dependent on the task
-or phenomenon studied. Here we follow the methodology proposed in [1]_ on the
+or phenomenon studied. Here we follow the methodology proposed in [1] on the
 openly available BCI IV 2a Dataset.
 
 
@@ -25,6 +25,17 @@
 Here, we use the augmentation module present in braindecode in the context of
 trialwise decoding with the BCI IV 2a dataset.
 
+References
+-----------
+
+.. [1] Rommel, C., Paillard, J., Moreau, T., & Gramfort, A. (2022)
+       Data augmentation for learning predictive models on EEG:
+       a systematic comparison. https://arxiv.org/abs/2206.14483
+
+.. [2] Banville, H., Chehab, O., Hyvärinen, A., Engemann, D. A., & Gramfort, A. (2021).
+       Uncovering the structure of clinical EEG signals with self-supervised learning.
+       Journal of Neural Engineering, 18(4), 046020.
+
 .. contents:: This example covers:
    :local:
    :depth: 2
@@ -109,7 +120,7 @@
 
 ######################################################################
 # Defining a list of transforms
-# --------------------
+# ------------------------------
 #
 # In this tutorial, we will use three categories of augmentations.
 # This categorization has been proposed by [1]_ to explain and aggregate
@@ -199,7 +210,7 @@
 if cuda:
     model.cuda()
 
-######################################################################
+##########################################################################
 # The model is now trained as in the trial-wise example. The
 # ``AugmentedDataLoader`` is used as the train iterator and the list of
 # transforms are passed as arguments.
@@ -233,7 +244,7 @@
 # generator of the training.
 
 train_X = SliceDataset(train_set, idx=0)
-train_y = array([y for y in SliceDataset(train_set, idx=1)])
+train_y = array(list(SliceDataset(train_set, idx=1)))
 
 #######################################################################
 #   Given the trialwise approach, here we use the KFold approach and
@@ -290,15 +301,3 @@
 eval_y = SliceDataset(eval_set, idx=1)
 score = search.score(eval_X, eval_y)
 print(f'Eval accuracy is {score * 100:.2f}%.')
-
-# References
-# ----------
-#
-# .. [1] Rommel, C., Paillard, J., Moreau, T., & Gramfort, A. (2022)
-#        Data augmentation for learning predictive models on EEG:
-#        a systematic comparison.
-#        https://arxiv.org/abs/2206.14483
-#
-# .. [2] Banville, H., Chehab, O., Hyvärinen, A., Engemann, D. A., & Gramfort, A. (2021).
-#        Uncovering the structure of clinical EEG signals with self-supervised learning.
-#        Journal of Neural Engineering, 18(4), 046020.

master/_downloads/995f58ef1b3d29dce64caee24e5bf457/plot_bcic_iv_2a_moabb_cropped.py

@@ -15,7 +15,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "\n# Searching the best data augmentation on BCIC IV 2a Dataset\n\nThis tutorial shows how to search data augmentations using braindecode.\nIndeed, it is known that the best augmentation to use often dependent on the task\nor phenomenon studied. Here we follow the methodology proposed in [1]_ on the\nopenly available BCI IV 2a Dataset.\n\n\n.. topic:: Data Augmentation\n\n    Data augmentation could be a step in training deep learning models.\n    For decoding brain signals, recent studies have shown that artificially\n    generating samples may increase the final performance of a deep learning model [1]_.\n    Other studies have shown that data augmentation can be used to cast\n    a self-supervised paradigm, presenting a more diverse\n    view of the data, both with pretext tasks and contrastive learning [2]_.\n\n\nBoth approaches demand an intense comparison to find the best fit with the data.\nThis view is supported by Rommel, C., Paillard, J., Moreau, T., & Gramfort, A. (2022),\nwho demonstrate the importance of the selection the right transformation and\nstrength for each different type of task considered.\nHere, we use the augmentation module present in braindecode in the context of\ntrialwise decoding with the BCI IV 2a dataset.\n   :depth: 2\n"
+        "\n# Searching the best data augmentation on BCIC IV 2a Dataset\n\nThis tutorial shows how to search data augmentations using braindecode.\nIndeed, it is known that the best augmentation to use often dependent on the task\nor phenomenon studied. Here we follow the methodology proposed in [1] on the\nopenly available BCI IV 2a Dataset.\n\n\n.. topic:: Data Augmentation\n\n    Data augmentation could be a step in training deep learning models.\n    For decoding brain signals, recent studies have shown that artificially\n    generating samples may increase the final performance of a deep learning model [1]_.\n    Other studies have shown that data augmentation can be used to cast\n    a self-supervised paradigm, presenting a more diverse\n    view of the data, both with pretext tasks and contrastive learning [2]_.\n\n\nBoth approaches demand an intense comparison to find the best fit with the data.\nThis view is supported by Rommel, C., Paillard, J., Moreau, T., & Gramfort, A. (2022),\nwho demonstrate the importance of the selection the right transformation and\nstrength for each different type of task considered.\nHere, we use the augmentation module present in braindecode in the context of\ntrialwise decoding with the BCI IV 2a dataset.\n\n## References\n\n.. [1] Rommel, C., Paillard, J., Moreau, T., & Gramfort, A. (2022)\n       Data augmentation for learning predictive models on EEG:\n       a systematic comparison. https://arxiv.org/abs/2206.14483\n\n.. [2] Banville, H., Chehab, O., Hyv\u00e4rinen, A., Engemann, D. A., & Gramfort, A. (2021).\n       Uncovering the structure of clinical EEG signals with self-supervised learning.\n       Journal of Neural Engineering, 18(4), 046020.\n   :depth: 2\n"
       ]
     },
     {
@@ -188,7 +188,7 @@
       },
       "outputs": [],
       "source": [
-        "train_X = SliceDataset(train_set, idx=0)\ntrain_y = array([y for y in SliceDataset(train_set, idx=1)])"
+        "train_X = SliceDataset(train_set, idx=0)\ntrain_y = array(list(SliceDataset(train_set, idx=1)))"
       ]
     },
     {
@@ -224,7 +224,7 @@
       },
       "outputs": [],
       "source": [
-        "import pandas as pd\nimport numpy as np\n\nsearch_results = pd.DataFrame(search.cv_results_)\n\nbest_run = search_results[search_results['rank_test_score'] == 1].squeeze()\nbest_aug = best_run['params']\nvalidation_score = np.around(best_run['mean_test_score'] * 100, 2).mean()\ntraining_score = np.around(best_run['mean_train_score'] * 100, 2).mean()\n\nreport_message = 'Best augmentation is saved in best_aug which gave a mean validation accuracy' + \\\n                 'of {}% (train accuracy of {}%).'.format(validation_score, training_score)\n\nprint(report_message)\n\neval_X = SliceDataset(eval_set, idx=0)\neval_y = SliceDataset(eval_set, idx=1)\nscore = search.score(eval_X, eval_y)\nprint(f'Eval accuracy is {score * 100:.2f}%.')\n\n# References\n# ----------\n#\n# .. [1] Rommel, C., Paillard, J., Moreau, T., & Gramfort, A. (2022)\n#        Data augmentation for learning predictive models on EEG:\n#        a systematic comparison.\n#        https://arxiv.org/abs/2206.14483\n#\n# .. [2] Banville, H., Chehab, O., Hyv\u00e4rinen, A., Engemann, D. A., & Gramfort, A. (2021).\n#        Uncovering the structure of clinical EEG signals with self-supervised learning.\n#        Journal of Neural Engineering, 18(4), 046020."
+        "import pandas as pd\nimport numpy as np\n\nsearch_results = pd.DataFrame(search.cv_results_)\n\nbest_run = search_results[search_results['rank_test_score'] == 1].squeeze()\nbest_aug = best_run['params']\nvalidation_score = np.around(best_run['mean_test_score'] * 100, 2).mean()\ntraining_score = np.around(best_run['mean_train_score'] * 100, 2).mean()\n\nreport_message = 'Best augmentation is saved in best_aug which gave a mean validation accuracy' + \\\n                 'of {}% (train accuracy of {}%).'.format(validation_score, training_score)\n\nprint(report_message)\n\neval_X = SliceDataset(eval_set, idx=0)\neval_y = SliceDataset(eval_set, idx=1)\nscore = search.score(eval_X, eval_y)\nprint(f'Eval accuracy is {score * 100:.2f}%.')"
       ]
     }
   ],

master/_downloads/9f3a9058fba292cd75e5ce23b492bc0c/plot_data_augmentation_search.py

@@ -293,11 +293,12 @@ def forward(self, x):
 
 ######################################################################
 # Training
-# --------
+# ---------
 #
 # We can now train our network on the pretext task. We use similar
-# hyperparameters as in [1]_, but reduce the number of epochs and increase the
-# learning rate to account for the smaller setting of this example.
+# hyperparameters as in [1]_, but reduce the number of epochs and
+# increase the learning rate to account for the smaller setting of
+# this example.
 
 import os