Closed tkgesis closed 1 month ago
When running benchmark_baseline.py with
feature_sets = ["fixation_number", "omission_rate", "reading_speed", 'sent_gaze', "mean_sacc_dur", "max_sacc_velocity", "mean_sacc_velocity", "max_sacc_dur", "max_sacc_amp", "mean_sacc_amp", 'sent_saccade', 'sent_gaze_sacc', "theta_mean", "alpha_mean", "beta_mean", "gamma_mean", "eeg_means", "sent_gaze_eeg_means", "electrode_features_theta", "electrode_features_alpha", "electrode_features_beta", "electrode_features_gamma", "electrode_features_all", "electrode_gaze_sacc"]
the variable 'sent' is not assigned for XBB
File "zuco-benchmark\src\benchmark_baseline.py", line 42, in get_or_extract_features
fe.extract_sentence_features(subject, f, feature_set, features, "")
File "zuco-benchmark\src\extract_features.py", line 221, in extract_sentence_features
weighted_nFix = np.array(af['duration']).shape[0] / len(sent.split())
UnboundLocalError: local variable 'sent' referenced before assignment
During processing of training data [Y..] , electrode_gaze_sacc IS NOT A VALID FEATURE SET.
appears several times.
Thank you for the recent commit. With the fixed versions, the script can be run partially. Windows filesystems won't handle file names with colons. To run the script on windows machines, one can change
datetime.now().strftime("%a, %d %b %Y %H%M%S"))
inWith the default config, the skript will extract features for the training data, but fails with the items from the heldout set.
OSError: Unable to open file (unable to open file: name = '../data/test/resultsXBB_NR.mat', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0)
According to https://osf.io/d7frw/wiki/home/ , the test data contain NR and TSR merged and shuffled, but the code expects two different files, one for each condition.
This could probably be fixed around
Hi, thank you for the notification! The paths for the heldout data in benchmark.py
should be updated now.
Hello, I'm sorry for the long wait time. You should now have access to all the features outlined in the configuration file. Additionally, we've updated the code for automatically generating submission files for each feature set. These files can be directly submitted to the challenge.
Thank you for the recent commit. With the fixed versions, the script can be run partially. Windows filesystems won't handle file names with colons. To run the script on windows machines, one can change
datetime.now().strftime("%a, %d %b %Y %H%M%S"))
in https://github.com/norahollenstein/zuco-benchmark/blob/33d03062fb4cf88e34dc7fd53cebf7c93d1cf86d/src/data_helpers.py#L20With the default config, the skript will extract features for the training data, but fails with the items from the heldout set.
According to https://osf.io/d7frw/wiki/home/ , the test data contain NR and TSR merged and shuffled, but the code expects two different files, one for each condition.
This could probably be fixed around https://github.com/norahollenstein/zuco-benchmark/blob/33d03062fb4cf88e34dc7fd53cebf7c93d1cf86d/src/benchmark.py#L25