The repository is to showcase how to develop deep learning algorithms also considering data versioning and model versioning. The model architecture used is VGG 16 with pre-trained weights. The execution steps were migrated from TensorFlow to Pytorch. More about the Tensorflow execution can be found in this link
Additional modifications were done to save the accuracy of the each class (cats and dogs).
- Create a virtual environment:
python3 -m venv .env - Activate the environment:
source .env/bin/activate - Install all the requirements :
pip install -r requirements.txt
- Download the data:
dvc get https://github.com/iterative/dataset-registry tutorials/versioning/data.zip - Unzip the data and remove the zip file:
unzip -q data.zip && rm -f data.zip - Register the data to DVC:
dvc add data - Train a model:
python train.py - Register the trained model to DVC:
dvc add checkpoints.pth - Commit all the changes and assign a version
- Download the next set of data:
dvc get https://github.com/iterative/dataset-registry tutorials/versioning/new-labels.zip - Follow the steps mentioned above from the second point.
The results observed for the first iteration are:
| train_acc | cats_train_acc | dogs_train_acc | test_acc | cats_test_acc | dogs_test_acc | |
|---|---|---|---|---|---|---|
| 0 | 95.57695595406967 | 96.50152348236993 | 94.65884451046051 | 97.25 | 98.5 | 96.0 |
The logs of the training done are listed below:
Generating Model
Populating Data
total_train_records: 1000
total_test_records: 800
Starting training
EPOCH: 1
Loss=1.431853175163269 Batch_id=99 Accuracy=88.50 Cats Accuracy=89.20 Dogs Accuracy=87.80: 100%|███████████████████████████████████████████████████████████████████████████████| 100/100 [00:20<00:00, 4.85it/s]
Test set: Average loss: 0.0350, Accuracy: 751/800 (93.88%) Cats Accuracy: 392/400 (98.00%) Dogs Accuracy: 359/400 (89.75%)
EPOCH: 2
Loss=0.006366188637912273 Batch_id=99 Accuracy=93.40 Cats Accuracy=93.00 Dogs Accuracy=93.80: 100%|████████████████████████████████████████████████████████████████████████████| 100/100 [00:19<00:00, 5.07it/s]
Test set: Average loss: 0.0279, Accuracy: 766/800 (95.75%) Cats Accuracy: 383/400 (95.75%) Dogs Accuracy: 383/400 (95.75%)
EPOCH: 3
Loss=0.8110159635543823 Batch_id=99 Accuracy=93.40 Cats Accuracy=93.20 Dogs Accuracy=93.60: 100%|██████████████████████████████████████████████████████████████████████████████| 100/100 [00:19<00:00, 5.04it/s]
Test set: Average loss: 0.0267, Accuracy: 761/800 (95.12%) Cats Accuracy: 383/400 (95.75%) Dogs Accuracy: 378/400 (94.50%)
EPOCH: 4
Loss=0.00015035287651699036 Batch_id=99 Accuracy=92.00 Cats Accuracy=92.40 Dogs Accuracy=91.60: 100%|██████████████████████████████████████████████████████████████████████████| 100/100 [00:19<00:00, 5.03it/s]
Test set: Average loss: 0.0531, Accuracy: 755/800 (94.38%) Cats Accuracy: 396/400 (99.00%) Dogs Accuracy: 359/400 (89.75%)
EPOCH: 5
Loss=3.576278118089249e-08 Batch_id=99 Accuracy=93.00 Cats Accuracy=93.20 Dogs Accuracy=92.80: 100%|███████████████████████████████████████████████████████████████████████████| 100/100 [00:20<00:00, 4.99it/s]
Test set: Average loss: 0.0353, Accuracy: 765/800 (95.62%) Cats Accuracy: 375/400 (93.75%) Dogs Accuracy: 390/400 (97.50%)
EPOCH: 6
Loss=7.152554815093026e-08 Batch_id=99 Accuracy=95.10 Cats Accuracy=94.40 Dogs Accuracy=95.80: 100%|███████████████████████████████████████████████████████████████████████████| 100/100 [00:20<00:00, 4.96it/s]
Test set: Average loss: 0.0823, Accuracy: 736/800 (92.00%) Cats Accuracy: 398/400 (99.50%) Dogs Accuracy: 338/400 (84.50%)
EPOCH: 7
Loss=1.382818254569429e-06 Batch_id=99 Accuracy=95.00 Cats Accuracy=96.40 Dogs Accuracy=93.60: 100%|███████████████████████████████████████████████████████████████████████████| 100/100 [00:20<00:00, 4.97it/s]
Test set: Average loss: 0.0241, Accuracy: 766/800 (95.75%) Cats Accuracy: 390/400 (97.50%) Dogs Accuracy: 376/400 (94.00%)
EPOCH: 8
Loss=0.0 Batch_id=99 Accuracy=95.50 Cats Accuracy=95.40 Dogs Accuracy=95.60: 100%|█████████████████████████████████████████████████████████████████████████████████████████████| 100/100 [00:20<00:00, 5.00it/s]
Test set: Average loss: 0.0312, Accuracy: 763/800 (95.38%) Cats Accuracy: 387/400 (96.75%) Dogs Accuracy: 376/400 (94.00%)
EPOCH: 9
Loss=4.1484063331154175e-06 Batch_id=99 Accuracy=95.30 Cats Accuracy=95.00 Dogs Accuracy=95.60: 100%|██████████████████████████████████████████████████████████████████████████| 100/100 [00:19<00:00, 5.01it/s]
Test set: Average loss: 0.0275, Accuracy: 766/800 (95.75%) Cats Accuracy: 389/400 (97.25%) Dogs Accuracy: 377/400 (94.25%)
EPOCH: 10
Loss=1.2568765878677368 Batch_id=99 Accuracy=96.40 Cats Accuracy=96.20 Dogs Accuracy=96.60: 100%|██████████████████████████████████████████████████████████████████████████████| 100/100 [00:19<00:00, 5.01it/s]
Test set: Average loss: 0.0321, Accuracy: 764/800 (95.50%) Cats Accuracy: 389/400 (97.25%) Dogs Accuracy: 375/400 (93.75%)The results observed for the second iteration are:
| train_acc | cats_train_acc | dogs_train_acc | test_acc | cats_test_acc | dogs_test_acc | |
|---|---|---|---|---|---|---|
| 0 | 95.89156880345085 | 96.5690201109141 | 95.10276442939261 | 95.375 | 96.5 | 94.25 |
The logs of the training done are listed below:
Generating Model
Populating Data
total_train_records: 2000
total_test_records: 800
Starting training
EPOCH: 1
Loss=0.021523717790842056 Batch_id=199 Accuracy=89.10 Cats Accuracy=88.80 Dogs Accuracy=89.40: 100%|███████████████████████| 200/200 [00:50<00:00, 3.99it/s]
Test set: Average loss: 0.0459, Accuracy: 754/800 (94.25%) Cats Accuracy: 395/400 (98.75%) Dogs Accuracy: 359/400 (89.75%)
EPOCH: 2
Loss=1.5382236242294312 Batch_id=199 Accuracy=93.15 Cats Accuracy=93.30 Dogs Accuracy=93.00: 100%|█████████████████████████| 200/200 [00:39<00:00, 5.04it/s]
Test set: Average loss: 0.0895, Accuracy: 726/800 (90.75%) Cats Accuracy: 399/400 (99.75%) Dogs Accuracy: 327/400 (81.75%)
EPOCH: 3
Loss=0.11554104089736938 Batch_id=199 Accuracy=92.05 Cats Accuracy=91.90 Dogs Accuracy=92.20: 100%|████████████████████████| 200/200 [00:39<00:00, 5.05it/s]
Test set: Average loss: 0.0402, Accuracy: 761/800 (95.12%) Cats Accuracy: 391/400 (97.75%) Dogs Accuracy: 370/400 (92.50%)
EPOCH: 4
Loss=0.017473753541707993 Batch_id=199 Accuracy=94.35 Cats Accuracy=94.30 Dogs Accuracy=94.40: 100%|███████████████████████| 200/200 [00:39<00:00, 5.05it/s]
Test set: Average loss: 0.0833, Accuracy: 751/800 (93.88%) Cats Accuracy: 397/400 (99.25%) Dogs Accuracy: 354/400 (88.50%)
EPOCH: 5
Loss=0.0 Batch_id=199 Accuracy=94.35 Cats Accuracy=94.70 Dogs Accuracy=94.00: 100%|████████████████████████████████████████| 200/200 [00:40<00:00, 4.97it/s]
Test set: Average loss: 0.0458, Accuracy: 764/800 (95.50%) Cats Accuracy: 381/400 (95.25%) Dogs Accuracy: 383/400 (95.75%)
EPOCH: 6
Loss=2.741809908002324e-07 Batch_id=199 Accuracy=93.15 Cats Accuracy=92.90 Dogs Accuracy=93.40: 100%|██████████████████████| 200/200 [00:40<00:00, 4.99it/s]
Test set: Average loss: 0.0400, Accuracy: 767/800 (95.88%) Cats Accuracy: 394/400 (98.50%) Dogs Accuracy: 373/400 (93.25%)
EPOCH: 7
Loss=1.3708603382110596 Batch_id=199 Accuracy=94.00 Cats Accuracy=94.10 Dogs Accuracy=93.90: 100%|█████████████████████████| 200/200 [00:40<00:00, 4.96it/s]
Test set: Average loss: 0.0313, Accuracy: 771/800 (96.38%) Cats Accuracy: 392/400 (98.00%) Dogs Accuracy: 379/400 (94.75%)
EPOCH: 8
Loss=3.576278118089249e-08 Batch_id=199 Accuracy=94.05 Cats Accuracy=94.50 Dogs Accuracy=93.60: 100%|██████████████████████| 200/200 [00:40<00:00, 4.93it/s]
Test set: Average loss: 0.0313, Accuracy: 765/800 (95.62%) Cats Accuracy: 384/400 (96.00%) Dogs Accuracy: 381/400 (95.25%)
EPOCH: 9
Loss=1.5497201388825488e-07 Batch_id=199 Accuracy=94.60 Cats Accuracy=94.50 Dogs Accuracy=94.70: 100%|█████████████████████| 200/200 [00:40<00:00, 4.93it/s]
Test set: Average loss: 0.0326, Accuracy: 764/800 (95.50%) Cats Accuracy: 389/400 (97.25%) Dogs Accuracy: 375/400 (93.75%)
EPOCH: 10
Loss=7.247662324516568e-06 Batch_id=199 Accuracy=95.35 Cats Accuracy=95.50 Dogs Accuracy=95.20: 100%|██████████████████████| 200/200 [00:40<00:00, 4.94it/s]
Test set: Average loss: 0.0309, Accuracy: 763/800 (95.38%) Cats Accuracy: 386/400 (96.50%) Dogs Accuracy: 377/400 (94.25%)