Embeddings with hoplite

[max-mauermann]

No description provided.

[kahst]

This is awesome, great work Max! A few minor things to (maybe) change:

• I would do a center crop of the search sample instead of first three seconds (any chance we can let users pick how to crop? maybe a slider below?)
• the default names of the embeddings database and (more importantly Dataset) could be "embeddings_database" and "my_dataset"
• can users specify a dataset in which to search or does search always include all datasets in a database?
• seems like no error message is shonw when something goes wrong. I got an error saying: ValueError: Embeddings [144 157 145 158 146 147 159 148 160 149 161 150 151 162 152 163 153 164 154 165 155 166 156 167] not found. and no results where retrieved. should at least be "something went wrong"
• the error happend after I added another dataset to the same database (it had 143 samples before, when loading, the GUI says 143 samples but it should have more when searching across datasets)

[kahst]

ok, it really seems that we can't search in databases with more than one dataset. The GUI only shows the sample count of the first database and search fails with an error similar to the one above (different indices).

[kahst]

everything else seems to work like a charm :)

[kahst]

One other thing I realized: When people set a band pass filter during embeddings extraction, is that being applied during search?

[Josef-Haupt]

Could we move that into gui-settings.json?

[max-mauermann]

no, the settings are per database and need to be saved along them. i will change this to use the metadata from the hoplite db, so no separate file is created.