diff --git a/README.md b/README.md index b88b363ae37389ff9efe856e9cacf6af42a681cb..eb5e7220f4653ed01c4e01c7e44c6e9a7cfa7deb 100644 --- a/README.md +++ b/README.md @@ -1,4 +1,25 @@ +The conDetSEC approach is implemented in the main_varyingLength.py file. +The Recurrent Neural Network autoencoder is implemented in the model.py file. + +The main_varyingLength.py file takes as input three information: +- The Directory in which data are stored +- The number of labelled samples to derive MUST and CANNOT Link constraints (this is used to access to the related fiile) +- The run identifier to add to the output file name + +In the data directory (i.e. JapWov) there are different files: +- data.npy: contains the data in numpy format as a tensor. (JapWov: (640, 29, 12) ) +- class.npy: contains the label assignment. (JapWov: (640,) ) +- seqLength.npy: contains the valid timestamps for each multi-variate time series contained in data.npy (JapWov: (640,)). The value of each row of this file is between 1 and the maximum lenght of the time series of the dataset + + Example to run the code: -python main.py JapWov 10 0 +python main_varyingLength.py JapWov 10 0 + +This means that we run the clustering algorithm on the Japanese Wovel dataset, with a number of constraints derived using 10 labelled instances per classes. The final 0 indicates the run number in order to record which is the specific run (for experimental evaluation purposes). +Running this line of code makes the hypothesis that in the JapWov folder, the file 10_0.npy exists. +The file 10_0.npy contains the identifier (position) of 10 samples per class that will be used to derive constraints. + +The "python main_varyingLength.py JapWov 10 0" line command will produce two files: +- JapWov/res_10_0.npy: The results of the K-Means clustering algorithm applied on the learnt embeddings +- JapWov/emb_10_0.npy: The data embedding produced by our framework -This means that we run the clustering algorithm on the Japanese Wovel dataset, with a number of constraints derived using 10 labelled instances per classes. The final 0 indicates the run number in order to record which is the specific run (for experimental evaluation purposes)