aliby.track.benchmark.CellBenchmarker¶
- class CellBenchmarker(meta, model, bak_model, nstepsback=None)[source]¶
Bases:
object
Takes a metadata dataframe and a model and estimates the prediction in a trap-wise manner.
This class can also produce confusion matrices for a given Tracker and validation dataset.
- Attributes
- masks
traps_loc
Generates a list of trap locations using the metadata.
Methods
Calculate all errors, addresses of images with errors and error fractions.
compare_traps
(exp, pos, trap)Error calculator for testing model and assignment heuristics.
gen_cm_stats
(pair[, thresh])Calculate confusion matrix for a pair of pos-timepoints
Calculates the trap-wise error and averages across a position.
Calculates the trap-wise error and averages across a position.
Requires self.meta
Predict all datasets defined in self.traps_loc
predict_set
(exp, pos, trap[, tp])Predict labels using tp1-tp2 accuracy of prediction
df_get_imglist
extract_pairs_from_trap
gen_cm_from_pairs
gen_pairlist
get_mota_stats
predict_lbls_from_tpimgs
Methods
__init__
(meta, model, bak_model[, nstepsback])Calculate all errors, addresses of images with errors and error fractions.
compare_traps
(exp, pos, trap)Error calculator for testing model and assignment heuristics.
df_get_imglist
(exp, pos, trap[, tp])extract_pairs_from_trap
(trap_loc)gen_cm_from_pairs
([thresh])gen_cm_stats
(pair[, thresh])Calculate confusion matrix for a pair of pos-timepoints
Calculates the trap-wise error and averages across a position.
gen_pairlist
()Calculates the trap-wise error and averages across a position.
get_mota_stats
(pair)Requires self.meta
Predict all datasets defined in self.traps_loc
predict_lbls_from_tpimgs
(tp_img_tuple)predict_set
(exp, pos, trap[, tp])Predict labels using tp1-tp2 accuracy of prediction
Attributes
masks
Generates a list of trap locations using the metadata.
- calculate_errsum()[source]¶
Calculate all errors, addresses of images with errors and error fractions.
- compare_traps(exp, pos, trap)[source]¶
Error calculator for testing model and assignment heuristics.
Uses the trap id to compare the amount of cells correctly predicted. This uses local indices, not whole timepoints. It returns the fraction of cells correctly predicted, and the timepoints of mistakes
Returns: float: Fraction of cells correctly predicted list of 2-sized tuples: list of tp id of errors and the mistaken cell
- gen_cm_stats(pair, thresh=0.7, *args, **kwargs)[source]¶
Calculate confusion matrix for a pair of pos-timepoints
- get_truth_matrix_from_pair(pair)[source]¶
Requires self.meta
args: :pair: tuple of size 4 (experimentID, position, trap (tp1, tp2))
returns
- Truth_mat
boolean ndarray of shape (ncells(tp1) x ncells(tp2) links cells in tp1 to cells in tp2
- property traps_loc¶
Generates a list of trap locations using the metadata.