[**PRACE AUTUMN SCHOOL 2021**](https://events.prace-ri.eu/event/1188/)
[**INTRODUCTION TO DEEP LEARNING**](https://hackmd.io/@pdl/Skn7I48MY)

# CLUSTER EXERCISES

## Exercise 1

### cifar10

| Submitter | Model description | Validation accuracy | Test accuracy |
| --------- | ----------------- | ------------------- | ------------- |
|Rajendra| Default parameters| 0.3863|
| Neha | 2 hidden layers = 50,50, drop rate= 0.001|42.77%|43.4%|
| Sreeram | 2 hidden layers = 50,30, drop rate= 0.001|41.96%|42.71%|
| Markus | Default | 0.3672 | 0.3693 |
|Febrian|lr:0.001-hidden1:50-hidden2:30-dropout:0.0-eval:True||41.70%|
|Greg|2 layers 100,100, drop rate 0.05, epochs 100|46.14%|44.8%|
| Markus | Mats's best result from TensorBoard | 0.4626 | |
|Rajendra| Hidden1 50 Hidden2 50 Dropout 0.001| 0.3939
|Rajendra| Hidden1 50 Hidden2 25 Dropout 0.2| 0.2363
|Rajendra| Hidden1 50 Hidden2 25 Dropout 0.1| 0.1870
|Ilja| --hidden1=2400 --hidden2=0 --dropout=0 --lr=0.001 | 0.4564 | 46.12% |
|Rajendra| Hidden1 50 Hidden2 40 Dropout 0.001| 0.4204
|Rajendra| Hidden1 50 Hidden2 40 Dropout 0.0| 0.4265

## Exercise 2

### dvc (dogs vs cats)

| Submitter | Model description | Test accuracy |
| --------- | ----------------- | ------------- |
| Markus | original dvc-cnn-simple.h5 | 73.03% |
| Markus | original dvc-vgg-reuse.h5 | 87.64% |
| Markus | original dvc-vgg-finetune.h5 | 91.52% |
| Febrian | original dvc-vgg-finetune.h5, 18 epoch | 92.34% |
| Neha | original dvc-vgg-finetune.h5 | 91.68%|
| Sreeram | original dvc-cnn-simple | 73.14% |
| Febrian | pytorch_dvc_cnn_pretrained.py, finetuned accuracy | 95.9% |

### gtsrb (traffic signs)

| Submitter | Model description | Test accuracy |
| --------- | ----------------- | ------------- |
| Markus | original gtsrb-cnn-simple.h5 | 72.58% |
| Markus | original gtsrb-vgg-reuse.h5 | 51.46% |
| Markus | original gtsrb-vgg-finetune.h5 | 62.38% |

## Exercise 3

| Submitter | Model description | Test accuracy | Runtime |
| --------- | ----------------- | ------------- | ------- |
| Markus | original tf2-20ng-rnn | 80.37% | 3:59 |
| Markus | original tf2-20ng-cnn | 87.05% | 1:18 |
| Rajendra | original tf2-20ng-cnn | 89.90% | |
| Sreeram | original tf2-20ng-cnn | 87.75% | 5:45 |
| Sreeram | original tf2-20ng-rnn | 81.05% | 2:01 |
|Neha| original tf2-20ng-cnn|87.58%| 6:58|
|Neha| original tf2-20ng-rnn|81.02%| 2:33|

## Exercise 4

Reporting only validation accuracy is fine, no need to run evaluate this time (unless you want to).

| Submitter | Model description | Val accuracy | Test accuracy | Runtime | No of GPUs |
| --------- | ------------------------------ | ------------ | ------------- | ---------- |------------|
| Rajendra | original tf2-dvc-cnn-simple.py | 0.7278 | | 0:00:25.81 | 1 |
| Rajendra | original tf2-dvc-cnn-simple-hvd.py |0.6966 | | |2 |
| Mats | tf2-dvc-cnn-simple with 2 GPUs | 0.7000 | | 0:00:40.75 | 2 |
| Mats | tf2-dvc-cnn-simple with 4 GPUs | 0.7098 | | 0:02:28.48 | 4 |
| Mats | tf2-dvc-cnn-pretrained (reuse only) with 1 GPU | 0.8982 | | 0:00:30.81 | 1 |
| Mats | tf2-dvc-cnn-pretrained (reuse only) with 2 GPUs | 0.8771 | | 0:00:29.41 | 2 |
| Sreeram | tf2-dvc-cnn-simple with mirrored strategy |0.7067| 62.38% |0:02:37.00|4|
| Neha | tf2-gtsrb-cnn-simple with mirrored strategy|0.8032||0:57:21.0|4|

## Exercise 5

| Submitter | Model description | Test accuracy | Runtime |
| --------- | ----------------- | ------------- | ------- |
| Markus | original tf2-20ng-rnn | 80.37% | 3:59 |
| Markus | original tf2-20ng-cnn | 87.05% | 1:18 |
| Mats | original tf2-20ng-bert| 83.30% | 14:13 |
| Rajendra | original tf2-20ng-cnn | 88.40% | |
| Rajendra | original tf2-20ng-rnn | 82.45% | |
| Rajendra | original tf2-20ng-bert| 82.90% |11.35 |
|Neha| original tf2-20ng-bert|82.55%| 0:11:29|
|Sreeram| original tf2-20ng-bert|81.95%| 0:10:08|
|Ilja| original tf2-20ng-bert|80.77%| 9 min|

## Exercise 6

**cifar10**

| Submitter | Hyperparameter values | Validation accuracy |
| --------- | --------------------- | ------------------- |
| Markus | 'epochs': 10, 'dropout': 0.0, 'lr': 0.001, 'hidden1': 495, 'hidden2': 46 | 0.4644 |
| Rajendra |epochs':10,'dropout':0.0,'lr':0.001, 'hidden1':296.89,'hidden2':103.68|0.4726|
|Murali|'dataset': 'cifar10', 'epochs': 10, 'dropout': 0.0, 'lr': 0.001, 'hidden1': 468.64888761686524, 'hidden2': 94.91400525177829|0.465|
| Markus | 'epochs': 10, 'dropout': 0.0, 'lr': 0.001, 'hidden1': 436, 'hidden2': 127 | 0.4757 |
|Neha|'dataset': 'cifar10', 'epochs': 10, 'dropout': 0.0, 'lr': 0.001, 'hidden1': 416.21516296827417, 'hidden2': 117.7365644492115|0.470200|
|Sreeram|'dataset': 'cifar10', 'epochs': 10, 'dropout': 0.0, 'lr': 0.0005, 'hidden1': 432.20447993164964, 'hidden2': 70.87518680387407|0.47139|
| ~ilja | {'dataset': 'cifar10', 'epochs': 10, 'dropout': 0.0, 'lr': 0.000130, 'hidden1': 273, 'hidden2': 902}| 0.507 |
| Markus | 'epochs': 10, 'dropout': 0.0, 'lr': 0.001, 'hidden1': 475, 'hidden2': 118 | 0.4756 |
| Febrian | {'dataset': 'cifar10', 'epochs': 10, 'dropout': 0.0, 'lr': 0.0005, 'hidden1': 465, 'hidden2': 124} | 0.492 |