--- a
+++ b/dnn_predicts/README.md
@@ -0,0 +1,46 @@
+# Model predictions on the test set
+This folder contain the deep neural network predictions on the test set. All files are in
+the format `.npy` and can be read using `numpy.load()`. Each one should contain a 
+
+All the content within this folder can be generate using the following sequence of commands:
+
+(without a GPU it should take about 25 minutes. With GPU acceleration it should take
+less then one minute)
+
+ ```bash
+cd /path/to/automatic-ecg-diagnosis
+PFOLDER="./dnn_predicts"
+MFOLDER="./model"
+DFOLDER="./data"
+
+# To generate the predictions on the test set corresponding to the main model used allong the paper use:
+
+python predict.py $DFOLDER/ecg_tracings.hdf5 $MFOLDER/model.hdf5 --output_file $PFOLDER/model.npy
+
+
+# We also train several networks with the same architecture and configuration
+# but with different initial seeds.  In order to generate the neural network 
+# prediction on the test set for each of these models:
+
+mkdir $FNAME/other_seeds
+for n in 1 2 3 4 5 6 7 8 9 10
+do
+python predict.py $DFOLDER/ecg_tracings.hdf5 $MFOLDER/other_seeds/model_$n.hdf5 --output_file $PFOLDER/other_seeds/model_$n.npy
+done
+
+
+# Finally, to asses the effect of how we structure our problem, we have considered alternative s
+# cenarios where we use 90\%-5\%-5\% splits, stratified randomly,
+# by patient or in chronological order. The predictions of those models in the test set
+# can be obtained using:
+
+mkdir $PFOLDER/other_splits
+for n in date_order individual_patients normal_order
+do
+python predict.py $DFOLDER/ecg_tracings.hdf5 $MFOLDER/other_splits/model_$n.hdf5 --output_file $PFOLDER/other_splits/model_$n.npy
+done
+```
+
+Where the `DFOLDER` should give the path to the folder containing the test dataset and `MFOLDER` should point to the 
+folder containing pre-trained models. The test dataset can be downloaded from [here](https://doi.org/10.5281/zenodo.3625006) and the
+pretrained models can be downloaded from here [here](https://doi.org/10.5281/zenodo.3625017)
\ No newline at end of file