https://github.com/holgerbrandl/yelp_restaurant_labels
https://github.com/holgerbrandl/yelp_restaurant_labels
Last synced: 8 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/holgerbrandl/yelp_restaurant_labels
- Owner: holgerbrandl
- Created: 2018-04-30T09:59:17.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2018-05-09T16:18:08.000Z (over 7 years ago)
- Last Synced: 2025-01-21T13:08:37.122Z (10 months ago)
- Language: Kotlin
- Size: 58.6 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
Yelp Restaurant Photo Classification with Kotlin and Deeplearning4j
===================================================================
For details see https://www.kaggle.com/c/yelp-restaurant-photo-classification
Download the data with
```
kaggle competitions download -c yelp-restaurant-photo-classification
```
References
* Other dl4j implementation (see gitter channel for discussion) https://gist.github.com/holgerbrandl/a522fa3156b05ec37fa55c117985c83a
## Remote execution
```bash
rsync --delete -avx --exclude target --exclude project ~/projects/deep_learning/kaggle_yelp_rest_pics/ brandl@talisker:~/projects/deep_learning/kaggle_yelp_rest_pics
```
```bash
screen -R yelp_kotlin
screen -R kotlin_yelp
cd ~/projects/deep_learning/kaggle_yelp_rest_pics
# http://www.gubatron.com/blog/2017/07/20/how-to-run-your-kotlin-gradle-built-app-from-the-command-line/
gradle run 2>&1 | tee yelp.$(date +'%Y%m%d').log
mailme "yelp done in $(pwd)"
```
## Misc
Data Prep
```bash
kaggle competitions download -c yelp-restaurant-photo-classification
#mkdir -p ~/projects/data/yelp-restaurant-photos
cd ~/projects/data/yelp-restaurant-photos
for tarFile in $(ls ~/.kaggle/competitions/yelp-restaurant-photo-classification/*.tgz); do tar xvf ${tarFile}; done
#rm train_photos/._*.jpg test_photos/._*.jpg
#https://stackoverflow.com/questions/11289551/argument-list-too-long-error-for-rm-cp-mv-commands
find train_photos -name "._*.jpg" -print0 | xargs -0 rm
find test_photos -name "._*.jpg" -print0 | xargs -0 rm
```
## Todo
What to use as base-line classifier vgg + custom dense layer?
Better use http://scikit-learn.org/stable/modules/generated/sklearn.naive_bayes.MultinomialNB.html as baseline (not random guessing https://machinelearningmastery.com/dont-use-random-guessing-as-your-baseline-classifier/)
E.g. use https://haifengl.github.io/smile/api/java/smile/classification/NaiveBayes.Model.html
Also see https://haifengl.github.io/smile/classification.html
Or even just Zero Rule (or 0R or ZeroR for short). It is the simplest rule you can use on a classification problem and it simply predicts the majority class in your dataset (e.g. the [mode](https://en.wikipedia.org/wiki/Mode_(statistics))).
http://scikit-learn.org/stable/modules/generated/sklearn.dummy.DummyClassifier.html
## References
https://tensorflow.rstudio.com/blog/keras-image-classification-on-small-datasets.html
* transfer learning increases accuracy from 80 to 90%