mhersche / eegnet-based-embedded-bci Goto Github PK

in trian_val_global_ss.ipynb

• I had to add np.transpose in all_acc[split_ctr] = np.transpose(np.genfromtxt(file, delimiter=' ')) in both plot_global and plot_ss functions
• the printed out global and ss in the ss experiment plot (last two cells) are only mean from fold 1. Correct code should be:
  global_acc = 100all_acc[:,:,1,0].mean()
  ss_acc = 100all_acc[:,:,1,-1].mean()

• You could add explicitly the fix for keras:

change "()" to "[ ]" in line no 1908,1910,1914, 1918 in anaconda3/envs/mybciEnv/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py (e.g. beta = tf.reshape(beta, [-1])) and have 'channels_last' in keras.json

• Remove sentence in models.py talking about channels_first

• get error when saving results "No such file or directory: 'results//stats/global_class_4_ds1_nch64_T3_split_0.csv'". suggestion: use os.path.join(). In save_results in main_global.py {experiment_name} missing: results_str = f'{results_dir}{experiment_name}/stats/global_class_{num_classes}_ds{n_ds}_nch{n_ch}_T{T}split{split_ctr}.csv' also line 139 in main_global.py