# DAY 2 EXERCISES

Select a free account from the table below and write your name to the "User" column. Passwords can be found in Zoom chat.

| Training account | | User |
| ---------------- | ----- | ------------------ |
| training085 | | Neha |
| training086 | | Sreeram |
| training087 | | Axel-Jan |
| training088 | | Purbaj |
| training089 | | Jose Manuel |
| training090 | | James Akash |
| training091 | | Samuel Boobier |
| training092 | | Aarne Klemetti |
| training093 | | Margherita |
| training094 | | Bruno |
| training095 | | Enrique |
| training096 | | Prakshal |
| training097 | | Prince |
| training099 | | Javed Razzaq |
| training098 | | Filip |
| training100 | | James Nelson |
| training101 | | Christian Meesters |
| training102 | | Kimmo Tikka |
| training103 | | Fabio Pratelli |
| training104 | | Ramkumar |
| training105 | | Narc F |
| training106 | | Trang |
| training107 | | Tiziano |
| training108 | | Michel |
| training109 | | |
| training110 | | Estev |
| training111 | | |
| training112 | | |
| training113 | | |
| training114 | | |
| training115 | | Sushanth Keshav |
| training116 | | |
| training117 | | Tomas F |
| training118 | | |
| training119 | | |
| training120 | | |
| training121 | | Jukka |
| training122 | | Victor T |
| training123 | | |
| training124 | | |
| training125 | | |
| training126 | | |
| training127 | | CVA |
| training128 | | |
| training129 | | Ayush Chaturvedi |
|training130 | |Marija |

## Exercise 5

| Dataset | Framework | Network architecture | Test accuracy | Submitter |
| ------- | --------- | --------------------------- | ------------- | ----------- |
| dvc | Keras | Simple CNN | 70.55% | Markus |
| dvc | Keras | VGG16 reuse | 88.61% | Markus |
| dvc | Keras | VGG16 finetuned | 92.41% | Markus |
| gtsrb | Keras | Simple CNN | 59.53% | Markus |
| gtsrb | Keras | VGG16 reuse | 42.95% | Mats |
| gtsrb | Keras | VGG16 finetuned | 62.40% | Mats |
| dvc | Keras | Simple CNN Scratch | 70.88% | Jose Manuel |
| dvc | Keras | Simple CN Scratch epochs=30 | 74.53% | Jose Manuel |
| dvc | Keras | Simple CNN Pretrained | 88.37% | Jose Manuel |
| dvc | Keras | Simple CNN Fine Tune | 92.88% | Jose Manuel |
| gtsrb | Keras | VGG16 Reuse | 45.27% | Jose Manuel |
| gtsrb | Keras | VGG16 finetuned | 63.33% | Jose Manuel |
| dvc | Keras | Simple CNN | 73.41% | Prakshal |
| dvc | Keras | VGG16 CNN Pretrained | 88.7% | Prakshal |
| dvc | Keras | VGG16 CNN Finetune | 92.62% | Prakshal |
| dvc | Keras | Simple CNN tfr | 69.79% | Prakshal |
| dvc | Keras | Simple CNN | 65.38% | Tomas |
| dvc | Keras | VGG16 pretrained | 88.19% | Tomas |
| dvc | Keras | Mobilnet Finetune | 96.90% | Tomas |
| dvc | Keras | Mobilnet Reuse | 96.51% | Tomas |
| dvc | Keras | Simple CNN | 69.05% | Margherita |
| dvc | Keras | Simple CNN Pretrained | 83.59% | Margherita |
| dvc | Keras | Simple CNN Finetune | 92.51% | Margherita |
| gtsrb | Keras | Simple CNN | 63.44% | Margherita |
| gtsrb | Keras | Simple CNN Pretrained | 46.13% | Margherita |
| gtsrb | Keras | Simple CNN Finetune | 60.68% | Margherita |
| dvc | Keras | simple cnn | 70.90% | Axel-Jan |
| dvc | Keras | simple cnn pretrained | 90.76% | Axel-Jan |
| dvc | Keras | Simple | 68.72% | Purbaj |
| dvc | Keras | Simple | 71.30% | Prince |
| dvc | Keras | Pre-trained | 85.53% | Prince |
| dvc | Keras | Fine tuned | 92.59% | Prince |
| gtsrb | Keras | Simple | 59.34% | Purbaj |
| dvc | Keras | Simple | 71.98% | Enrique |
| dvc | Keras | VGG16 reuse | 88.04% | Sreeram |
| dvc | Keras | VGG16 finetuned | 91.99% | Sreeram |
| dvc | Keras | VGG16 reuse | 87.90% | Neha |
| dvc | Keras | VGG16 finetuned | 92.47% | Neha |
| dvc | Keras | Pre-Trained | 92.29% | Purbaj |
| gtsrb | Keras | Pre-Trained | 62.64% | Purbaj
|dvc | Keras | Simple | 72.4% | James
| dvc | Keras | Simple | 73.20% | Kimmo
| dvc | Keras | reuse | 83.41% | Kimmo
| dvc | Keras | finetune |  92.28% | Kimmo
| dvc | PyTorch | simple cnn | 67.39% | Javed |

---

## Exercise 6

| Framework | Network architecture | Test accuracy | Runtime | Submitter |
| --------- | ---------------------- | ------------- | -------------------------------------- | ----------- |
| Keras | Default RNN | 78.22% | 2:59 | Markus |
| Keras | Default CNN | 85.30% | 0:50 | Mats |
| Keras | Default BERT | 81.60% | 15:09 | Mats |
| Keras | RNN | 82.63% | 02:55 | Kimmo |
| Keras | CNN | 87.02% | 03:38 | Kimmo |
| Keras | CNN w 20 epochs | 94.88% | 01:10 (?) | Kimmo |
| Keras | BERT | 82.92% | 11:30 | Kimmo |
| Keras | RNN | 76.65% | 04:48 | Jose Manuel |
| Keras | CNN | 88.42% | 02:50 | Jose Manuel |
| Keras | BERT | 82.07% | 12:03 | Jose Manuel |
| Keras | RNN Adamax n_epochs=30 | 55.37% | 03:43 | Jose Manuel |
| Keras | RNN n_epochs=30 | 90.50% | 04:45 | Jose Manuel |
| Keras | CNN | 86.7% | 00:52 | Prakshal |
| Keras | RNN | 80.55% | 04:03 | Prakshal |
| Keras | BERT | 81.83% | 11:19 | Prakshal |
| Keras | RNN | 82.42% | 2:47 | Tomas |
| Keras | CNN | 86.33% | 1:09 | Tomas |
| Keras | BERT | 82.55% | 9:33 | Tomas |
| Keras | CNN | 89.15% | 0:56 (sacct -l - j JOBID --> cpu time) | NFF |
| Keras | RNN | 79.80% | 2:41 (same) | NFF |
| Keras | Bert | 83.25% | 9:19 (same) | NFF |
| Keras | RNN | 80.77% | 3:11 | Purbaj |
| Keras | CNN | 87.60 | 1:45 | Purbaj |
| Keras | CNN | 82.53 | 11.04 | Sreeram |
| Keras | CNN with epochs=50 | 96.82% | 2.04 | Sreeram |
| Keras | RNN | 78.57% | 11:16 |Neha |
| Keras | CNN | 86.37% | 02:42 | Neha |
| Keras | BERT | 83.10% | 00:50 | Neha |
| Keras | CNN | 88.10% | | Axel-Jan |
| Keras | RNN | 77.93% | | Axel-Jan |
| Keras | BERT | 82.53% | | Axel-Jan |
| Keras | CNN | 87.12% | 2:08 | Margherita |
| Keras | RNN | 82.40% | 4:17 | Margherita |
| Keras | BERT | 83.17% | 11:25 | Margherita |
| Keras | BERT | 82.38 | 12:29 | Purbaj |
| Keras | CNN epoch =100 | 96.75% | 3:44 (seff) | NFF |
| Keras | RNN | 82.88% | 2:46 | Bruno |
| Keras | CNN | 86.87% | 0:51 | Bruno |
| Keras | BERT | 82.70% | 11:18 | Bruno |
| Keras | CNN | 90.18% | 0:51 | Enrique |
| Keras | RNN | 78.68% | 2.58 | Enrique |
| Pytorch | BERT | 81.68% | 24.58 | Ayush |
| Pytorch | CNN | 73.68% | 4.58 | Ayush |
| Pytorch | CNN batchsize=32,epoch=50 | 73.68% | 2.58 | Ayush |
| Pytorch | RNN | 66.68% | 24.58 | Ayush |
| Keras | BERT | 80.83% | 1:52:30 | Enrique |
|KERAS|CNN|2:44|88.27%|Marija|

## Exercise 7

| Framework | Network | Approach | GPUs | Test accuracy | Runtime | Submitter |
| --------- | -------------------------- | ---------------- | ---- | ------------- | ------- | ----------- |
| Keras | Simple CNN | Baseline | 1 | 70.75% | 0:36 | Mats |
| Keras | Simple CNN | MirroredStrategy | 4 | 69.12% | 1:03 | Mats |
| PyTorch | Simple CNN | DataParallel | 4 | ? | 3:42 | Mats |
| Keras | Simple CNN | Horovod | 2 | 64.53% | 0:34 | Mats |
| Keras | Simple CNN | MirroredStrategy | 2 | 73.98% | 1:46 | Jose Manuel |
| Keras | Simple CNN | MirroredStrategy | 4 | 68.84% | 2:44 | Jose Manuel |
| Keras | Simple CNN | Horovod | 4 | 70.53% | 3:25 | Sreeram |
| Keras | Simple CNN | Horovod | 4 | 65.80% | 3:25 | Neha |
| Keras | Simple CNN | MirroredStrategy | 2 | 71.82% | 1:39 | Javed |