Moved excess information and documentation to external dev doc #55826
Conversation
…into tyrone/hoc2023-ai-documentation
| # Math explanation: Cosine distance outputs a value between 0 -> 1 where smaller values = greater similarity | ||
| # We can redefine this into cosine similarity with a simple (x-1)*-1 due to their mathematical relationship | ||
| # Since we take the SUM(MAX(similarity)) value when determining which options to present the user, cosine similarity is preferable | ||
| # Conversion from cosine distance to cosine similarity for easier readability in frontend computations. |
There was a problem hiding this comment.
Nit: front-end since used as adjective. (Sorry I didn't include in prior comment.)
| # Conversion from cosine distance to cosine similarity for easier readability in frontend computations. | ||
| # Math explanation: Cosine distance outputs a value between 0 -> 1 where smaller values = greater similarity. | ||
| # Cosine similarity redefines this relationship so that instead larger values = greater ssimilarity. | ||
| # Since we expose some of these values to students in the frontend, we felt that similarity values growing larger would be easier to understand. |
There was a problem hiding this comment.
Nit suggestion: 'Since we expose some of these values to users in the frontend, we felt that greater values = greater similarity would be easier to understand.'
| # Since we take the SUM(MAX(similarity)) value when determining which options to present the user, cosine similarity is preferable | ||
| # Conversion from cosine distance to cosine similarity for easier readability in frontend computations. | ||
| # Math explanation: Cosine distance outputs a value between 0 -> 1 where smaller values = greater similarity. | ||
| # Cosine similarity redefines this relationship so that instead larger values = greater ssimilarity. |
There was a problem hiding this comment.
Nit: typo 'similarity'
| @@ -7,7 +7,7 @@ Run the `HoC2023AiGenerateWeights.py` to generate the associated output weights | |||
|
|
|||
| Before running the script, make sure to adjust your local parameters based off the current model being used. Previous iterations have leveraged spaCy and OpenAI's Ada models and it is not unreasonable to anticipate that the model "vendor" may change again in the future. | |||
|
|
|||
| As of 01/08/2024, this script uses AWS's Titan v1 LLM through their Bedrock API. | |||
| As of 01/08/2024, this script uses AWS's Titan v1 LLM through their Bedrock API. For additional background context and testing resources, check the google drive here: https://docs.google.com/document/d/1beDoalfB1Y7BybN82YGhuzTNos_TE5l0dX4XdKPzNdw/edit?usp=sharing | |||
There was a problem hiding this comment.
You could link text to URL as done in javabuilder READAME
'For additional background context and testing resources, see the Dance AI Design Dev Doc.'
There was a problem hiding this comment.
Also - could you define 'LLM' the first time it's introduced in the doc?
| measure the relatedness of text strings. The distance between two vectors measures | ||
| their relatedness. https://developers.google.com/machine-learning/crash-course/embeddings/video-lecture | ||
| These embeddings are stored in caches files as pickle files, python's native way to serialize data. | ||
| The embeddings used to generate these maps are stored in cached pickle files such as foreground_embeddings.pkl to prevent duplicate LLM API calls. |
There was a problem hiding this comment.
Nit: use backticks for file name:
'The embeddings used to generate these maps are stored in cached pickle files such as foreground_embeddings.pkl to prevent duplicate LLM API calls.'
|
|
||
| At runtime, DanceAI will use the three maps to lookup the scores for each output type and take the top three indexes of (MAX(SUM(Input1Scores, Input2Scores, Input3Scores))) to select a final palette/foreground/background to display to the user. These maps are stored as a local cache rather than generated at runtime to remove the costs associated with querying a LLM and improve runtime performance. | ||
| At runtime, DanceAI will use the three maps to lookup the scores for each output type and randomly select one of the top 3 results of MAX(SUM(Input1Scores, Input2Scores, Input3Scores)) to select a final palette/foreground/background to display to the user. These maps are stored as a local cache rather than generated at runtime to remove the costs associated with querying a LLM and improve runtime performance. |
There was a problem hiding this comment.
Since on the frontend, we present both randomly selected top version (from top 3) and randomly selected bottom versions ( 4 from bottom 20), maybe we can just state that the three maps are used to look up scores for each output type. We could also refer them to the calculateOutputSummedWeights.ts file.
…tation Moved excess information and documentation to external dev doc
The below is a follow-up fix for the conversation in this PR: #55633
Executive Summary