lfisher7/crmd-classifier

CRMD-classifier

Predicting cardiac rhythm management devices (CRMDs) manufacturers using deep learning

Scripts:

Data processing:

• number_of_models - plots bar chart of number of each class in test and train data.
• mean_image - takes the mean of images within given folder.
• std_image - takes the SD of images within given folder.
• data_preprocessing - applies histogram equalisation to images.
• image_resizing - resizes images in given path to 227x227
• image_to_numpy - creates numpy arrays of data with size 150x150 or 227x227.

Model building:

• first model - builds and trains simple model.
• first_model_binary - builds and trains binary classification model.
• export_features - runs images through convolutional base and saves output
• load_features - takes output from export features and trains classifier
• InceptionV3_scratch - trains model with InceptionV3 architecture - not initialise on ImageNet.
• scratch_hyperas - building a model from scratch using hyper wrapper for grid search.
• conv_base_frozen - transfer learning using ImageNet weights.
• fine_tuning_final_run - similar to conv_base frozen, but fine tunes top layers.
• fold_creator - splits data into 10 folds for cross validation
• fold_directory - trains model with fine-tuning and cross-validation.

Data analysis:

• plot_tensorboard - takes tensor board training traces and plots using plotly.
• predicting_confusion - plots confusion matrices of predictions on test set.

Final programme:

• CRMD_predictor - makes a prediction of CRMD model on images stored in images_to_run folder.