Project PP 2022

• Mihai-Valentin DUMITRU
• Vlad-Andrei BĂDOIU
• Teodora-Andreea ION
• Mihai UDUBAȘA
• Matei POPOVICI

• Your Teaching Assistant
  • You are highly encouraged to talk with your TA about the project and to ask questions during lab. You may also have sessions outside lab hours to check your code with them (either requested by you or by them).

• You can get a maximum of 4.2 points for the project, over the entire semester.
• You need a minimum of 2.5 points from the project in order to qualify for the exam, along with a minimum of 3 points accumulated during the semester (lab, project, lecture).
• Your final score will be computed at the end of the semester.
• There will be automated testing, but the score given by the checker is not necessarily the final score. There can be points removed by either:
  1. not complying to the task requirements (e.g. using library functions instead of implementing a required function), in which case the task is considered not completed and the points for that task are removed.
  2. not meeting a deadline (see below).

• The project will be divided in 3 steps, each with its own task set, submission assignment and number of points associated. The task sets will have progressively-increasing difficulty, and will reflect the lecture progress.

• Each task set has its own deadline; there is also a hard deadline for the project, on the 15th of May.
• You can submit solutions to each taskset until the project deadline.
• If you miss a task set deadline, you will lose 0.7 points from that taskset (to a minimum of zero).
• To meet a task set deadline, you have to make a submission worth at least 0.7 points by that deadline; the grading team might manually deduce some points after the manual review, but unless this is done for extreme circumstances (e.g. hardcoding answers to automated tests), it will not count as missing the deadline.

We are working for the Haskell Health company that builds fitness trackers. For this project we want to build a basic data science analysis framework similar to pandas from Python.

The exercises are based on and tested with the tables that you can found in the file Dataset.hs:

• emails: contains the connection between the name of a person and their email
• eight_hours: contains the number of steps during 8 hours on a specific day
• physical_activity: contains the total steps, total distance, very active minutes, fairly active minutes, lightly active minutes of a person on a specific day
• sleep_min: contains the number of minutes slept in a week

When working on the project, we suggest parsing the table cells in a single function and passing appropriate data types to helper functions (e.g. for an eight_hours table, a helper function might work on a (String, [Float])).

During the whole project we will consider the following types: type CSV = String type Value = String type Row = [Value] type Table = [Row]

At the beginning of the Tasks.hs file you already have implemented the functions split_by, read_csv, write_csv that you can use to solve the tasks but you can also implement your own.

Important: Use the function (printf “%.2f”) when you have to transform a Float number to String.

Your module with the solution will have to be called Tasks.hs.

We offer you an automated checker and a test suite. We also encourage you to add your own tests.

To run the checker, simply run runhaskell main.hs in your shell (if you only wish to run certain tasksets, you can give it additional arguments: runhaskell main.hs 1.4 2.3 2.4).

The checker is written in Haskell and can be easily extended to add new testcases. Check main.hs for more.

The solution must be submitted to vmchecker. You need to upload a zip archive containing the file Tasks.hs as well as any other source files used. The archive should also contain a README file.

Deadline is on the 10th of April, 23:55

We will first like to compute everyone's average number of steps per given day, based on the formula: $(10+11+12+13+14+15+16+17)/8$. You will have to write a function which takes an eight_hours table and returns a table with columns “Name”, “Average Number of Steps”.

Based on their number of steps, check:

• how many people have achieved their daily goal (have at least 1000 steps given the total number of steps on the 8 hours from the eight_hours table),
• what is the percentage of people who have achieved their goal and
• what is the average number of daily steps for all people.

For this you will have to implement 3 functions which take an eight_hours table and return an Int.

We want to find out which are the hardest hours to be active. We want to compute the average steps for each hour. The result will be a table with columns “H10”, “H11”, “H12”, “H13”, “H14”, “H15”, “H16”, “H17”, where the “Hx” column will represent the average number of steps for the x hour in the eight_hours table.

Similar to the previous task, we want to know how many minutes were spent being active based on their intensity. You will use the physical_activity table and compute a table with 3 rows “VeryActiveMinutes”, “FairlyActiveMinutes”, “LightlyActiveMinutes” and 3 columns (“range1”, “range2”, “range3”), each column with a range of number of minutes: first column will represent the range $[0-50)$, second $[50-100)$ and third $[100,500)$. A value in the table found at the intersection between a column and a row will be the number of people who have spent a number of minutes included in the range given by the column at the intensity given by the row.

At this point, we would like to see the overall leaderboard. You'll have to sort the people by their total number of steps (ascending). If two people have the same number of steps, they will be listed in alphabetical name order. The function for this will take a physical_activity table and return a table with columns “Name”, “Total steps”.

Hint: Use the Ordering data type

Then we will compute the difference between two parts of the day regarding the number of steps based on the table eight_hours. We will create a table with 4 columns: “Name”, “Average first 4h”, “Average last 4h”, “Difference”. This table will contain the people sorted ascending by the “Difference” column. If 2 people have the same difference, they will be listed in alphabetical name order.

Implement a function which applies a function to all values in a table.

vmap :: (Value -> Value) -> Table -> Table

An example use of this would be:

correct_table = vmap (\x -> if x == "" then "0" else x) table

Implement a function which applies a function to all entries (rows) in a table. The new column names are given. Additionally, you have to implement a function which takes a Row from sleep_min table and returns a Row with 2 values: email, total number of minutes slept that week.

rmap :: (Row -> Row) -> [String] -> Table -> Table
get_sleep_total :: Row -> Row

For the get_sleep_total function, print the number of minutes using (printf “%.2f”).

Deadline is on the 28th of April, 23:55

Write a function which takes a column name and a Table and returns the Table sorted by that column (if multiple entries have the same values, then it is sorted by the first column). Sorting is ascending for columns with numeric values and lexicographical for strings. If the value is missing (“”), then it is considered before all values (e.g. “” < “0”).

tsort :: String -> Table -> Table

Implement a function which takes Tables t1 and t2 and adds all rows from t2 at the end of t1, if column names coincide. If columns names are not the same, t1 remains unchanged.

vunion :: Table -> Table -> Table

Implement a function which takes Tables t1 and t2 and extends each row of t1 with a row of t2 (simple horizontal union of 2 tables). If one of the tables has more lines than the other, the shorter table is padded with rows containing only empty strings.

hunion :: Table -> Table -> Table

Implement table join with respect to a key (a column name). If an entry from table 1 has the same value for the key as an entry from table 2, their contents must be merged. If there are other columns with the same name in both tables, then the value from t2 overrides the value from t1, unless it is null (“”).

tjoin :: String -> Table -> Table -> Table

Write the implementation for the cartesian product of 2 tables. The new column names are given. To generate the entries in the resulting table you have to apply the given operation on each entry in t1 with each entry in t2.

cartesian :: (Row -> Row -> Row) -> [String] -> Table -> Table -> Table

Extract from a table those columns specified by name in the first argument.

projection :: [String] -> Table -> Table

Filter rows based on a given condition applied to a specified column.

filterTable :: (Value -> Bool) -> String -> Table -> Table

Deadline is on the 12th of May, 23:55

For the last part of our work at our data science framework at Haskell Health we will need two important functionalities: a query language and graph queries. This will make enable the user of our framework to quickly develop a product based on our framework. Moreover, to showcase the power of our framework to the marketing team, we will do typo corrections on the data using a distance function.

The objective of this task set is to:

• allow programmers to combine table processing functions in a flexible way
• create a separation between what functions do and how they are implemented. This is helpful for a number of reasons:
  • integration of new table functionality later on
  • continuous code upgrade (new algorithms for different table processing functions can be implemented without altering the rest of the project)
  • debugging and testing

You will implement a query language which represents a variety of table transformations, some of which you have already implemented as table functions. A Query is a sequence (in some cases, a combination) of such transformations. You will also implement an evaluation function which performs a Query on a given table.

The query language to be implemented is shown below:

data Query =
    FromTable Table
    | AsList String Query
    | Sort String Query
    | ValueMap (Value -> Value) Query
    | RowMap (Row -> Row) [String] Query
    | VUnion Query Query
    | HUnion Query Query
    | TableJoin String Query Query
    | Cartesian (Row -> Row -> Row) [String] Query Query
    | Projection [String] Query
    | forall a. FEval a => Filter (FilterCondition a) Query
    | Graph EdgeOp Query
 
-- where EdgeOp is defined:
type EdgeOp = Row -> Row -> Maybe Value
 

Don't worry about Graph or Filter queries yet.

While most queries take a Table and return a transformed Table, there are some queries which evaluate to a list of String. Thus we define the type: QResult which describes any of these two possible query result types:

data QResult = Table Table | List [String]

In the skeleton we enroll QResult in class Show.

Task 3.1.: In order to ensure separation between queries and their evaluation we define class Eval, which offers function eval. Your job is to enroll Query in this class.

class Eval a where
    eval :: a -> QResult
 
instance Eval Query where
    ...

We explain below how each data constructor from Query should be evaluated:

1. FromTable table: returns the table it received as a parameter. Basically doesn't do anything. We do this such that we can pass tables as queries.
2. AsList colname query: returns values from column colname as a list.
3. Sort colname query: sorts table by column colname.
4. ValueMap op query: maps all values from table, using op.
5. RowMap op colnames query: maps all rows from table, using op.
6. VUnion query1 query2: vertical union of the 2 tables obtained through the evaluations of query1 and query2.
7. HUnion query1 query2: horizontal union of the 2 tables.
8. TableJoin colname query1 query2: table join with respect to column colname.
9. Cartesian op colnames query1 query2: cartesian product.
10. Projection colnames query: extract specified columns from table.

You can look into the reference file to better understand how these functions work.

You may have noticed that filter query is commented-out. You can uncomment it at this stage. Filtering will receive a special treatment. Because filter conditions are usually complex, instead of performing successive filter queries it is better to build complex query conditions. For this reason, we define type FilterCondition a, illustrated below:

data FilterCondition a =
    Eq String a |
    Lt String a |
    Gt String a |
    In String [a] |
    FNot (FilterCondition a) |
    FieldEq String String

Remark: the type FilterCondition a is polymorphic because such conditions may be expressed over (in this homework) two types:

• Float and
• String

We briefly explain what each condition expresses:

1. Eq colname ref: checks if value from column colname is equal to ref.
2. Lt colname ref: checks if value from column colname is less than ref.
3. Gt colname ref: checks if value from column colname is greater than ref.
4. In colname list: checks if value from column colname is in list.
5. FNot cond: negates condition.
6. FieldEq colname1 colname2: checks if values from columns colname1 and colname2 are equal.

A FilterCondition must evaluate to an actual filtering function, which has type:

type FilterOp = Row -> Bool

Since such filtering functions work differently for FilterCondition Float and FilterCondition String, we need a class FEval which contains function feval. The latter is used to evaluate a FilterCondition a to a function of type FilterOp. In order to do so, feval needs to have information about column names (the table head), hence it's type is shown below.

class FEval a where
    feval :: [String] -> (FilterCondition a) -> FilterOp

Task 3.2.: Your task is to write the instances for (FEval Float) and (FEval String).

instance FEval Float where
    ...
instance FEval String where
    ...

Task 3.3.: Now you can write the evaluation for the data constructor Filter query (see function eval from the previous section).

A graph is a special kind of table which has precisely the following column names: [“From”, “To”, “Value”]. Each row defines a weighted edge between node From and node To.

The query Graph edgeop query: creates such a table starting from the result of the evaluation of query. Suppose the query evaluates to a table T.

• The nodes are the rows in table T.
• The weight of an edge between 2 nodes is given by edgeop, which returns a Maybe Value. If edgeop row1 row2 returns Nothing, then we don't have an edge between those 2 nodes. If it returns Just val then we have an edge between row1 and row2 of weight val.
• In the resulting table, each row describes an edge between node_i and node_j and will have the values:
  • “From” = first column from node_i
  • “To” = first column from node_j
  • “Value” = edgeop node_i node_j

The edge node_i-node_j is the same as node_j-node_i, so it should only appear once (graphs are unoriented). “From” value should be lexicographically before “To”.

Example: Suppose T is the table shown below:

If we would like to build a graph that connects all users in the same category (e.g. 321 is sports people), then:

edgeop [_,_,z] [_,_,c]
   | z == c = Just c
   | otherwise = Nothing

and the resulting graph will be:

If we would like to build a graph that connects users with hours of sleep equal or with a difference of at least a point, then:

edgeop [_,x,_] [_,y,_]
   | abs $ (read x :: Int) - (read y :: Int) <= 1 = Just "similar"
   | otherwise = Nothing

and the resulting graph is:

Task 3.4.: Implement eval for Graph queries.

We want to check the similarities between users’ hours slept.

• For that, we want to obtain a graph where “From” and “To” are users’ emails and “Value” is the distance between the 2 users’ hours slept.
• We define the distance between user1 and user2 as “the sum of intervals where they both slept a equal amount” (e.g. same value for TotalMinutesAsleep1). Keep only the rows with distance >= 5.
• The edges in the resulting graph (the rows in the resulting table) should be sorted by the “Value” column. If email is missing, don't include that entry.

Task 3.5.: Your task is to write similarities_query as a sequence of queries, that once evaluated results in the graph described above.

Note: similarities_query is a Query. The checker applies eval on it.

1. Enroll Query in class Eval (without Filter or Graph). 0.3p
2. Enroll FilterCondition in class FEval and implement eval for Filter query. 0.2p
3. Implement eval for Graph query. 0.2p
4. Extract similarity graph. 0.3p

For the final tasks of this project, we want to merge the information about users’ stats from the 3 tables (8h of steps, minutes sleep week, physical activity in a day). The issue with joining these 3 tables is that table Minutes_slept_week is mapped using users’ emails, whereas the other 2 tables are mapped using users’ names.

Fortunately, we have another table, email_map, that maps users’ names to their emails. The bad news is that this table contains typos.

Your job is to fix the typos in email_map table, generate the correct table and then use that to join the information about user’s stats.

Column Name from table email_map contains typos. Example: Zane Kayle is misspelled as Zan Kayle or Amelia Caden is misspelled as Amelia Camdden. Fortunately, the name Zane Kayle contains only a few misspelled letters and it is guaranteed to appear in the correct form, in the name column of the 8h sleep or physical_activity_in_a_day tables. Hence, we can use the latter names as reference to correct the typos.

The goal is to write function correct_table, which receives the name of the column containing typos, the table containing typos, a reference table (a table where the values from that column are correct, but not necessarily in the same order) and returns the first table where the typos have been fixed. All parameters are strings in CSV-format.

correct_table :: String -> CSV -> CSV -> CSV
-- this will be tested using: correct_table "Nume" email_map_csv hw_grades_csv

So in order to fix the typos, you will use a reference table, where the specified column has correct values.

• extract the necessary column from the table with typos and the reference table (let's call these T and Ref);
• filter out only the values from T and Ref which don't have a perfect match in the other table - these are the problematic entries (this will help improve time performance);
• calculate the distance between each value from T and each value from Ref (distance = how similar the 2 strings are - you decide how to formally define this distance);
• for every value from T, its correct form is the value from Ref with the shortest distance to it;
• lastly, restore the original table, replacing the incorrect values from T with the correct values from Ref.

• Your implementation must be generic! It can't depend on the table structure or rows order. You can't assume that the rows in the the faulty table and in the reference table have the same order. Similarly, you can't choose a distance function that only works for these tables.
• The steps above are only recommended. If you find another implementation that works and respects the first condition (generic implementation), that's great!
• Also recommended is that you use your previously implemented Query Language, but it's not a restriction. You might find some queries really helpful, such as Cartesian, Projection or Filter.