Assignment 4: MAIS202: Kaggle Competition

MAIS202 Winter2021 Assignment 4: Find the maximum digit in an image

We implemented the VGG16 architecture for our model. Here is how you can reproduce our results

Run the jupyter notebook from this repo or Run the trained model by downloading it from this link

Steps to run the Notebook from this Repo

1. Initial steps

Running the first cell will download all the data required.

Run cell 2 to import all the required libraries.

Run cell 3 to load the data as numpy arrays.

Run cell 4 to mount to your google drive.

Run cell 5 to print some images from the training set.

Finally run cell 6 to load the training images and cells under “Preparing the data” to prep the dataset.

2. Implementing the model

Run the cell under “Defining the model” to define our implementation of the VGG16 architecture.

3. Training

Run all the cells under “Training” to train the model and save it to your drive.

4. Testing

Run the cells under “Making inferences on test set” to test the model.

This will save the test results to a csv file on your drive. Download it to see the results!

Steps to run the trained model from the provided link

1. Download it

!gdown --id 1Ns4Y8ibEQFHZkEBtZrLvbSMzWNWG3TRx

2. Load it in you notebook using

from model import MNISTClassifier

model = pt.load(path_to_the_downloaded_model)

3. Load the testing images using

!wget -O test_x.npy  https://www.dropbox.com/s/qfbaw6a18cthkg4/test_x.npy?dl=0
test_images = np.load("test_x.npy")

4. Make inferences by calling the inference method as follows:

from utils import inference

inference(model, array_of_128x128_images, path_to_save_predictions.csv)

Note that the model can only operate on $128 \times 128$ grayscale images.

5. Downlad the csv files with the results!