GovTech Internship 2023, GDS-CC4 submission, FOO CHUAN SHAO

Move to the parent directory and install all Python dependencies with pip3 install -r requirements.txt.

From the parent directory, move to the tasks directory with cd tasks.
Run tasks using the command of python task#.py, filling up "#" with the desired task (1-3)
(e.g. python task1.py will run task 1)

Alternatively, you can run this with any code editors using the "Run" button, or its equivalent.

Similar to running tasks, however, move to the tests directory with cd tests followed by python -m unittest for all tests.
Alternatively for individual tests, python -m unittest tests.test#, replacing "#" with the desired task (1-3)

A simple way of deploying this app would be to deploy it with a serverless framework as Function as a Service (FaaS) on service providers such as Google Cloud Function or AWS Lambda.

However, without changes, the app currently is inflexible in the data that it can take in. Therefore, some (minimal) work will need to be done to allow the function to take in arguments such as the ability to read from a URL or a JSON object. Other than this, the application would be ready to be deployed while satisfying specifically the tasks assigned.

The current functions could also be adapted to serve a broader utility purpose. For example, in task 2, there could be arguments taking in the start and end date, creating a search function, which only requires a small change in in_april_2019.

The benefits of creating this as a Cloud Function would be that it can be a microservice. There could be a larger encompassing GUI that calls on the function, making use of the functionalities of the code. This could also have benefits such as being easy to maintain and debug since it is small by itself.

Alternatively, the code could be integrated and built into an end-to-end service. This will require more functionalities such as add frontend, backend, security aspects. However, all of these will require much more time and effort, which includes but is not limited to creating databases, designing a GUI using Flask or other frameworks, and more. These applications can then be deployed on services such as GCloud, AWS, Heroku, or any service of choice.

In my opinion I would not recommend building it as an end-to-end service unless if the GUI provides more services than simply these tasks, otherwise there would be too much overhead to simply deploy.

In conclusion, these tasks are considerably specific and limited in its range, thus, an ideal way to deploy this application would be as a Cloud Function, making it easy to use and easy to maintain.

1. (Task 1) Invalid Country-Code is regarded as missing data and is populated as NA
2. Missing data keys are replaced with "NA" instead of discarding the entire row
3. Leading and trailing whitespaces is to be removed
4. JSON data will always have the same structure and will have the same keys for a particular value
5. (Task 2) The duration of the event in April 2019 does not matter, as long as it has occurred for at least a second in April 2019.
6. (Task 3) Only exact matches in rating_text shall be considered.

1. Some of the data may have special characters that can not be read by Python by default. To fix, encoding to 'utf-8' is required.
2. The number of results_found and results_shown does not correctly represent the number of restaurants found, thus should not be considered in verification in the tasks.
3. (Task 1) Strange data for all country_id of 17, where the city is dummy and the ratings are 0. To remove the data, uncomment line 28.
4. (Task 2) There may be more than 1 event photos, so they are populated in a list of photos.
5. (Task 3) Some of the rating_text are in another language other than English. However, translating these data may result in it becoming valid. To include these data, uncomment lines 26 to 40.

Task 1 simply reads through a JSON file and navigating through it using the appropriate list traversal and dictionary keys. The below demonstrates the methods and reasoning behind the design of the code.

1. Pandas is used for easy .csv and .xlsx management.
2. .get method is used for missing data keys as per stated in assumption

Task 2 is similar to task 1 in getting the necessary information. However, it includes a condition to ensure that event is in April 2019.
As per the condition, there are 3 particular cases where the event will happen on April 2019.\

1. If the start date is in April 2019
2. If the end date is in April 2019
3. If the event starts BEFORE April 2019 AND ends AFTER April 2019

Task 3 can have multiple approaches to calculating the threshold for the various ratings. The approach taken in the code is to simply pair the rating_text and the average corresponding rating_score. Other approaches may include machine learning.

Additionally, there seems to be different languages for rating_text, which makes it difficult to translate. However, these texts when translated might give the expected keys for the various ratings. By assumption 6, these ratings are not considered even if the translation gives the expected results.
However, as per comment 4, the consideration has also been coded in and will need to be uncommented to include these ratings.

The output of this task will be printed on the terminal, otherwise the main() function will return the dictionary of the data.
For the output printed on the terminal, the score will be rounded to 1 decimal place.

Test 1 tests for the flexibility of the code while dealing with JSON keys and missing values. This is the basic test to cover to ensure that the code works as expected.

In this test, 2 files have been specifically modified to simulate the cases for missing keys and empty values. The expected outcome of the modified file is then used and asserted to match the actual outcome.

Test 1 can be further enhanced with more edge cases using fuzzy testing.

Test 2 handles the functionality and ensures the edge cases of the dates. There are a total of 6 cases that was tested and asserted if the event has occured in April 2019 or not.

In this test, missing keys and empty values has been omitted in the test due to time constraint. However, to do so, the same approach to test 1 can be taken to test comprehensively.

Test 3 ensures that the calculation for the average score has been done correctly and appropriately. For this test, the approach taken is to edit the file and expect for an outcome.

The simplest and easiest test to create was to edit the rating for "Poor" since it has only one count. This also helped to test edge cases, and managed to fix 1 bug regarding the division of 0.