feature_reader#

class analysis.feature_reader.FeatureReader(feature_dir: str | PathLike, feature_file: str | PathLike = '', binarize_label: bool = True)[source]#
static filter_features(feature_columns: list, ch_name: str | None = None, list_feature_keywords: list[str] | None = None) → list[source]#

filters read features by ch_name and/or modality

Parameters:

  • feature_columns (list)

  • ch_name (str, optional)

  • list_feature_keywords (list[str], optional) – list of feature strings that need to be in the columns, by default None

Returns:

features – column list that suffice the ch_name and list_feature_keywords

Return type:

list

static get_epochs(data, y_, epoch_len, sfreq, threshold=0) → tuple[ndarray, ndarray][source]#

Return epoched data.

Parameters:

  • data (np.ndarray) – array of extracted features of shape (n_samples, n_channels, n_features)

  • y (np.ndarray) – array of labels e.g. ones for movement and zeros for no movement or baseline corr. rotameter data

  • epoch_len (int) – length of epoch in seconds

  • sfreq (int/float) – sampling frequency of data

  • threshold (int/float) – (Optional) threshold to be used for identifying events (default=0 for y_tr with only ones and zeros)

Returns:

  • epoch_ (np.ndarray) – array of epoched ieeg data with shape (epochs,samples,channels,features)

  • y_arr (np.ndarray) – array of epoched event label data with shape (epochs,samples)

normalize_features() → DataFrame[source]#

Normalize feature_arr feature columns

Returns:

z-scored feature_arr

Return type:

pd.DataFrame

plot_all_features(ch_used: str | None = None, time_limit_low_s: float | None = None, time_limit_high_s: float | None = None, normalize: bool = True, save: bool = False, title='all_feature_plt.pdf', ytick_labelsize: int = 10, clim_low: float | None = None, clim_high: float | None = None)[source]#

_summary_

Parameters:

  • ch_used (str, optional)

  • time_limit_low_s (float, optional)

  • time_limit_high_s (float, optional)

  • normalize (bool, optional)

  • save (bool, optional)

  • title (str, optional) – default “all_feature_plt.pdf”

  • ytick_labelsize (int, optional) – by default 10

  • clim_low (float, optional) – by default None

  • clim_high (float, optional) – by default None

plot_cort_projection() → None[source]#

_summary_

plot_subject_grid_ch_performance(subject_name=None, performance_dict=None, plt_grid=False, feature_str_add='performance_allch_allgrid')[source]#

plot subject specific performance for individual channeal and optional grid points

Parameters:

  • subject_name (string, optional) – used subject, by default None

  • performance_dict (dict, optional) – by default None

  • plt_grid (bool, optional) – True to plot grid performances, by default False

  • feature_str_add (string, optional) – figure output_name

plot_target_averaged_channel(ch: str = '', list_feature_keywords: list[str] | None = None, features_to_plt: list | None = None, epoch_len: int = 4, threshold: float = 0.1, normalize_data: bool = True, show_plot: bool = True, title: str = 'Movement aligned features', ytick_labelsize=None, figsize_x: float = 8, figsize_y: float = 8) → None[source]#

_summary_

Parameters:

  • ch (str, optional)

  • list_feature_keywords (Optional[list[str]], optional)

  • features_to_plt (list, optional)

  • epoch_len (int, optional)

  • threshold (float, optional)

  • normalize_data (bool, optional)

  • show_plot (bool, optional)

  • title (str, optional) – by default “Movement aligned features”

  • ytick_labelsize (_type_, optional)

  • figsize_x (float, optional) – by default 8

  • figsize_y (float, optional) – by default 8

plot_target_avg_all_channels(ch_names_ECOG=None, list_feature_keywords: list[str] = ['stft'], epoch_len: int = 4, threshold: float = 0.1)[source]#

Wrapper that call plot_features_per_channel for every given ECoG channel

Parameters:

  • ch_names_ECOG (list, optional) – list of ECoG channel to plot features for, by default None

  • list_feature_keywords (list[str], optional) – keywords to plot, by default [“stft”]

  • epoch_len (int, optional) – epoch length in seconds, by default 4

  • threshold (float, optional) – threshold for event detection, by default 0.1

read_results(performance_dict: dict = {}, subject_name: str | None = None, DEFAULT_PERFORMANCE: float = 0.5, read_grid_points: bool = True, read_channels: bool = True, read_all_combined: bool = False, ML_model_name: str = 'LM', read_mov_detection_rates: bool = False, read_bay_opt_params: bool = False, read_mrmr: bool = False, model_save: bool = False, save_results: bool = False, PATH_OUT: str = '', folder_name: str = '', str_add: str = '')[source]#

Save performances of a given patient into performance_dict from saved nm_decoder

Parameters:

  • performance_dict (dictionary) – dictionary including decoding performances, by default dictionary

  • subject_name (string, optional) – subject name, by default None

  • DEFAULT_PERFORMANCE (float, optional) – chance performance, by default 0.5

  • read_grid_points (bool, optional) – true if grid point performances are read, by default True

  • read_channels (bool, optional) – true if channels performances are read, by default True

  • read_all_combined (bool, optional) – true if all combined channel performances are read, by default False

  • ML_model_name (str, optional) – machine learning model name, by default ‘LM’

  • read_mov_detection_rates (boolean, by defaulte False) – if True, read movement detection rates, as well as fpr’s and tpr’s

  • read_bay_opt_params (boolean, by default False)

  • read_mrmr (boolean, by default False)

  • model_save (boolean, by default False)

  • save_results (boolean, by default False)

  • PATH_OUT (string, by default None)

  • folder_name (string, by default None)

  • str_add (string, by default None)

Returns:

performance_dict

Return type:

dictionary

static read_target_ch(feature_arr: DataFrame, label_name: str, binarize: bool = True, binarize_th: float = 0.3) → ndarray[source]#

_summary_

Parameters:

  • feature_arr (pd.DataFrame) – _description_

  • label_name (str) – _description_

  • binarize (bool, optional) – _description_, by default True

  • binarize_th (float, optional) – _description_, by default 0.3

Returns:

_description_

Return type:

_type_

run_ML_model(feature_file: str | None = None, estimate_gridpoints: bool = False, estimate_channels: bool = True, estimate_all_channels_combined: bool = False, output_name: str = 'LM', save_results: bool = True)[source]#

machine learning model evaluation for ECoG strip channels and/or grid points

Parameters:

  • feature_file (string, optional)

  • estimate_gridpoints (bool, optional) – run ML analysis for grid points, by default True

  • estimate_channels (bool, optional) – run ML analysis for ECoG strip channel, by default True

  • estimate_all_channels_combined (bool, optional) – run ML analysis features of all channels concatenated, by default False

  • model (sklearn model, optional) – ML model, needs to obtain fit and predict functions, by default linear_model.LogisticRegression(class_weight=”balanced”)

  • eval_method (sklearn.metrics, optional) – evaluation performance metric, by default metrics.balanced_accuracy_score

  • cv_method (sklearn.model_selection, optional) – valdation strategy, by default model_selection.KFold(n_splits=3, shuffle=False)

  • output_name (str, optional) – saving name, by default “LM”

  • save_results (boolean) – if true, save model._coef trained coefficients