WatsonInnovation

Repository for the Watson Innovation course project

Requirements

In order to use this code, the following requirements must be satisfied:

1. Java JDK 7 or higher
2. Python 2.7 and Python 3
3. os, argparse, time, sys, urllib2, and json Python modules (installable via pip)

How to use

Currently, Pearl does not have an user interface. Consequently, it is possible to explore its features with .jar file with command lines options.

Firstly, you need to download all the files from the following link

Secondly, you need to open the config.csv file and insert there the credentials (separated from their identifier by one tab) for the following services:

• Watson Visual Recognition
• Watson Natural Language Understanding
• Watson Discovery (and Environment & Collection identifiers)

Finally, you need to edit the europeana-client.properties file and add your API key for Europeana.

Now you are ready to use the jar file. You can start the program launching the command java -jar Backend.jar. Once it is running, the console will ask you if you want to perform a training of the models (option t) or a benchmarking of your models (option b). Obviously, to perform a benchmark it is necessary to train the model first - which might take up to one hour.

During the training process, the application will save all the information about paintings in the res/smartdb.csv file and all the images downloaded from Google in the images/ directory. Finally, in the json/ folder there will be all the informations related to the painings saved in json format, ready to be processed by Watson Discovery (reminder: due to API limitations, the files need to be uploaded manually). In the file res/similarItems.csv you can find similarity scores (i.e. cosine distance) for each pair of paintings.

If you want to perform a benchmark of Discovery, you can do it by checking how the service performs answering the queries contained in the res/queries.csv file, checking the answer provided in the folder queriesResults. On the other hand, if you want to benchmark the Visual Recognition service, you can choose via console the subfolder contained in res/benchmark/ that holds the images to classify. The Visual Recognition results are saved in the res/benchmark.csv and res/wrongPredictions.csv files.

Have fun!