ca_exp_eval
Description
Helper functions for evaluation of standard concept embedding analysis experiments.
Experiment results are assumed to have a folder structure of
<root>/layers/<layer_id>/<timestamp>/<concept_name>
.
For details see the respective sample Sacred experiment scripts.
Attributes
Layer IDs of pytorch AlexNet model in correct order. |
|
Layer IDs of pytorch VGG16 model in correct order. |
|
Layer IDs of pytorch Mask R-CNN model in correct order. |
Functions
|
Get the analysis root folder path for given |
|
Provide a nice display style of |
|
Gather dicts with all_stats, best_emb_stats, and merged_stats for different roots. |
|
Gather a DataFrame with the stats of all best embeddings. |
|
Gather a DataFrame with the stats of all embedding runs. |
|
Best embedding for given layer. |
|
Stats of the best embeddings of an analysis as DataFrame. |
|
Get all the concept names available for ALL layers in the given experiment |
|
Get all the concept names available for the given experiment root. |
|
Embeddings of an analysis as series indexed by |
|
Return all available layers for the given analysis root. |
|
Stats of an analysis run as DataFrame. |
|
Get the saved visualization of the best embedding. |
|
Highlight the maximum in a Series with red font color. |
|
Merge the stats from all runs and from the best embeddings to an overview over the test values of |
|
Provide a plot of metric values against layers for one concept and all settings in |
|
Provide a plot of metric values against layers for all concepts in |
|
Plot and save the output of |
|
Plot and save the output of |