udacity / aihcnd_c2_starter Goto Github PK

Hi,

What is an acceptable AUC score? The paper mentioned in this README.md file is of ChestNext Paper, obviously we can't compare our results with their results due to lack of infrastructure. The AUC for pneumonia images is worst than random guessing for all the model architectures that I have been trying like VGG-16, VGG-19, Densenet121!

Hi,

There is a discrepancy in the files present on Kaggle website and in this repository.

sample_labels.csv contains only 5606 records while the original records are 112120

The Data_Entry_2017.csv file in the workspace contains an extra column called Unnamed: 11
This column does not contain any useful data and is also not present in the original dataset. Then why was this column added?

all_xray_df = pd.read_csv('/data/Data_Entry_2017.csv')
all_image_paths = {os.path.basename(x): x for x in
glob(os.path.join('/data','images*', '', '.png'))}
print('Scans found:', len(all_image_paths), ', Total Headers', all_xray_df.shape[0])
all_xray_df['path'] = all_xray_df['Image Index'].map(all_image_paths.get)
all_xray_df.sample(3)

The following code in EDA.ipynb references data folder which is not present in the github repo.

Hi,
As described on Readme file, there is no Image folder for patient images to perform training or EDA. Are we supposed to find the data somewhere else? If so, can you provide us with mode details to where to get the data from?

Thanks

HI,

The correct source of data must be :

https://nihcc.app.box.com/v/ChestXray-NIHCC

This website contains the newest available data with train test splits already!