Comments (2)
Quickly answering now, as I won't be available for next 2 weeks.
During training we noted that the denominator reported in the progress bar (10,800) and the duration of training (~7 hours) for the initial epoch (see screenshot below) was considerably more than for the subsequent epoches (101 through 106), where the denominator was 1,350 and duration of training was less than 1 hour each.
The progress bar you sent is the scheduler of the learning rate. The model is gonna be ran for a few (fake) epochs at the beginning of the training (and every 14 epochs) to choose an "optimal" learning rate. This is decorrelated from the epoch duration that you can observe after.
This was a little confusing given that it appears 6 separates models were used for prediction (which aligns to our 6 sets of model weights resulting from our training) however our understanding was that the validate step would select the single best performing model and apply that alone. So we are unclear why the 'apply' command would assess all 6 models, and assuming this is the case, we see once again a dramatic difference in duration of time required for this assessment between the first interation and subsequent ones (9min:45sec for the first versus less than 2 minutes for the rest).
For inference time :
First iteration : the model is applied on the validation set.
Then the thresholds (computed during validation) for each of the classes are applied (those are the next 5 iterations).
From what I can see, everything seems normal. You can have a look at the files generated during the validation/inference time to see if anything seems weird (the onset/offset parameters for instance).
from voice-type-classifier.
Thank you for the clarification!
from voice-type-classifier.
Related Issues (20)
- KeyError: 'Domain' when retraining model HOT 4
- ./apply.sh does not work for numeric uris HOT 1
- CUDA out of memory error during validation HOT 3
- Differences in electronic speech HOT 4
- Fine-tuning versus Retraining HOT 2
- ./apply.sh: Something went wrong when applying the model HOT 5
- something went wrong when applying th emodel HOT 4
- Conda virtual env error made with the vtc.yml file HOT 2
- windows users!!! HOT 6
- citation unclear in README, arxiv paper doesn't contain info on accepted paper HOT 1
- CondaEnvException: Pip failed when creating "pyannote" HOT 2
- np.long issue HOT 1
- Something went wrong when applying the model HOT 1
- problem validating after retraining (err: ALICE/21) HOT 4
- pip install numba fails on M1 macs HOT 1
- Issue when applying model HOT 9
- Strange distribution of duration of outputted vocalisations HOT 3
- ImportError: cannot import name 'Pretrained' from 'pyannote.audio.features HOT 2
- Issue with self-recursion, fresh install on a new machine. HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from voice-type-classifier.