Skip to content
This repository has been archived by the owner on Sep 18, 2023. It is now read-only.

Latest commit

 

History

History
175 lines (121 loc) · 5.62 KB

README.md

File metadata and controls

175 lines (121 loc) · 5.62 KB
Raw

Mentor Pipeline

Use these tools to create/update a mentor for Mentorpal.

Quick Start

Required Setup (session videos and timstamps)

TODO: needs to download session videos and timestamp files automatically but for now... Download session videos and timestamps and place them under ./data/mentors/{mentor_id}/build/recordings eg.

data/mentors/my_new_mentor/
├── build
│   ├── recordings
│   │   ├── session1
│   │   │   ├── p001-bio-long.csv
│   │   │   ├── p001-bio-long.mp4
│   │   │   ├── p002-some-utterances.csv
│   │   │   ├── p002-some-utterances.mp4
│   │   ├── session2
│   │   │   ├── p001-more-questions.csv
│   │   │   ├── p001-more-questions.mp4

Configure AWS Credentials

TODO:

  • figure out a better way to handle AWS credentials for running locally
  • add more specific details about required permissions for the AWS IAM

The pipeline uses AWS Transcribe, and for now, you MUST set your aws credentials in the env, e.g. either export them to your shell

export AWS_ACCESS_KEY_ID=<your iam id>
export AWS_SECRET_ACCESS_KEY==<your iam key>

...or just pass both of the above to every make call.

Build and Test a Mentor

If raw video, audio and timestamp files for a mentor are stored in S3 (more on this below), we can use the following commands to build a classifier for the mentor. Note that videos are not required to generate a classifier to a new mentor.

Create/update the training data
make data/mentors-{mentor_id}
Train {mentor} classifier
make checkpoint/{mentor_id}

Run mentorpal cluster with {mentor} data and classifier

cd .. && make local-run-dev

Generate web videos (if you're ready to run the website and not just classifier)

make videos/mentors/{mentor_id}

Example endpoints for mentorpal cluster

  • Mentor Homepage (Test web site) http://localhost:8080/mentorpanel/?mentor={mentor}
  • Mentor API (Test classifier, e.g. with Postman) http://localhost:8080/mentor-api/questions/?mentor={mentor}&query={question}

Pipeline Overview

The classification data pipeline can be used to create all data needed for a usable mentor from raw recording files.

Output

After running the pipeline the following files will be generated:

  • {mentor}/data/classifier_data.csv
  • {mentor}/data/topics.csv
  • {mentor}/data/utterance_data.csv
  • videos/{mentor}/mobile/idle.mp4
  • videos/{mentor}/web/idle.mp4
  • videos/{mentor}/mobile/{mentor}_{video_id}.mp4
  • videos/{mentor}/web/{mentor}_{video_id}.mp4

Supplementary Documentation

Generating Timestamp Files

After the interview is done, watch it fully and note down the start and end timestamps for each question. Timestamp files should be in a CSV file of the following format:

Notes Answer/Utterance Question Response start Response end
(string) (char: A/U) (string) (timestamp HH:MM:SS) (timestamp HH:MM:SS)

DEV

Make/ENV variables for DEV

When you're working on the mentor-pipeline tool set, you frequently want to run make rules with a local build of the mentor-pipeline docker image and/or local copies of the python source for its python modules. The Makefile has a number of variables to support local development.

NOTE

All the examples below prepend variables to a make call, but you can also always export any of these variables once and they will stay in effect for the remainder of your shell session, e.g.

export DOCKER_IMAGE=mentor-pipeline:latest
# then later ...
make shell

...instead of

DOCKER_IMAGE=mentor-pipeline:latest make shell

DOCKER_IMAGE

Change the mentor-pipeline docker image from the current published release, e.g.

Change the docker image for a single make call like this

DOCKER_IMAGE=mentor-pipeline:latest make shell

...or configure it for your shell session like this

export DOCKER_IMAGE=mentor-pipeline:latest

DEV_ENABLED

Set DEV_ENABLED to have make rules run with local source for python modules. Will only use local sources for specific modules if the source is found at default (or configured) paths (details below)

DEV_ENABLED=1 make shell

DEV_ROOT

A default root for all python modules in dev. If you have set DEV_ENABLED=1 and any of the python modules listed below are cloned there, they will automatically be includes. The default value for DEV_ROOT is ~/projects

DEV_MENTOR_PIPELINE

Override the path to where mentor-pipeline is cloned. NOTE source will only be used if DEV_ENABLED=1

DEV_TRANSCRIBE

Override the path to where py-transcribe is cloned. NOTE source will only be used if DEV_ENABLED=1

DEV_TRANSCRIBE_AWS

Override the path to where py-transcribe-aws is cloned. NOTE source will only be used if DEV_ENABLED=1

Running mentor-pipeline docker shell

You can open a shell to the pipeline docker image like this:

make shell

All dev variables described above will apply to the shell.

Once in the docker shell, you can run the pipeline script directly, e.g.

python mentor_pipeline_runner.py --mentor some_mentor_id --data-update --data=/app/mounts/data/mentors