MENU
A table in the Lakehouse named customer_churn_params is used in churn prediction by the machine learning team. The table contains information about customers derived from a number of upstream sources. Currently, the data engineering team populates this table nightly by overwriting the table with the current valid values derived from upstream data sources.
The churn prediction model used by the ML team is fairly stable in production. The team is only interested in making predictions on records that have changed in the past 24 hours.
Which approach would simplify the identification of these changed records?
Suggested Answer: D
Get Premium Databricks Certified Data Engineer Professional Questions
as Interactive Web-Based Practice Test or PDF
Real Exam Format
Detailed Explanations
Instant Access
Get Your Practice Test
Delta Lake, built on top of Parquet, enhances query performance through data skipping, which is based on the statistics collected for each file in a table. For tables with a large number of columns, Delta Lake by default collects and stores statistics only for the first 32 columns. These statistics include min/max values and null counts, which are used to optimize query execution by skipping irrelevant data files. When dealing with highly nested JSON structures, understanding this behavior is crucial for schema design, especially when determining which fields should be flattened or prioritized in the table structure to leverage data skipping efficiently for performance optimization. Reference: Databricks documentation on Delta Lake optimization techniques, including data skipping and statistics collection (https://docs.databricks.com/delta/optimizations/index.html).
47 comments · newest first
AS
Ashlyn
10 months ago
B could be interesting, but is it really necessary to switch to streaming?
SU
Sue
10 months ago
Wait, does E really handle all the edge cases?
IZ
Izetta
10 months ago
Not sure about D, seems like it could complicate things.
LA
Lashawna
11 months ago
I agree, C is efficient for only predicting on changed records.
LE
Lezlie
11 months ago
Option C seems like a solid approach for tracking changes.
LU
Luke
11 months ago
I feel like option C makes sense because it focuses on comparing previous predictions, but I’m not sure if that’s the simplest approach compared to others.
AM
Amber
11 months ago
I practiced a similar question where we had to decide between batch and streaming processing; I wonder if option B could be a good fit since it talks about incrementally predicting.
MI
Micheal
12 months ago
I think option D sounds familiar because it involves adding a timestamp, which could help track changes, but I’m not entirely confident if it’s the most efficient way.
CO
Cory
12 months ago
I remember discussing how using a change data capture approach could help identify only the records that have changed, but I'm not sure which option aligns best with that.
RI
Ricki
12 months ago
Option D seems like a straightforward solution. Adding a timestamp field and using that to filter the data could be a simple and effective way to solve this problem. I'll need to consider the pros and cons, but it's definitely an option worth exploring.
CR
Cruz
1 year ago
I'm leaning towards option E. Replacing the overwrite logic with a merge statement and using a change data feed seems like a good way to identify the changed records. That way, we can focus the predictions on just the updated data.
LI
Ling
1 year ago
Option B sounds interesting, using Structured Streaming to incrementally process the data. That could be a more efficient approach than re-processing everything each time. I'll need to think through the details, but it seems like a promising solution.
KA
Katlyn
1 year ago
I'm a bit confused by the question. It's not clear to me how the different options would work in practice. I might try to get some clarification from the instructor before attempting to answer.
LI
Lilli
1 year ago
This seems like a tricky question. I'm not sure if I fully understand the requirements, but I think option C might be a good approach. Calculating the difference between previous predictions and the current data could help identify the changed records.
BO
Bobbye
1 year ago
Option D seems like a simple solution, but I'm not sure if it's the most efficient. Adding a timestamp field and then using that to identify the records written on a particular date could work, but it might not be as scalable as some of the other options.
CY
Cyril
1 year ago
I'm leaning towards option E. Replacing the overwrite logic with a merge statement and then making predictions on the changed records seems like a straightforward way to handle this. It's a common pattern, so I feel pretty confident about that approach.
KA
Kanisha
1 year ago
Option B with Structured Streaming sounds like an interesting approach. It could help us process the data more efficiently and only make predictions on the records that have changed. I'll have to look into that a bit more.
HU
Huey
1 year ago
I'm a bit confused by this question. It's not entirely clear to me how the different options would work in practice. I might need to think through the pros and cons of each approach a bit more.
CR
Crissy
1 year ago
This seems like a tricky question, but I think option C might be the way to go. Calculating the difference between the previous predictions and the current data could help us focus on just the customers that have changed.
AM
Amos
1 year ago
Option B sounds like the data engineering team's dream come true. Structured Streaming? Sounds like they just want to play with the latest and greatest tech.
LA
Lakeesha
1 year ago
E) Replace the current overwrite logic with a merge statement to modify only those records that have changed; write logic to make predictions on the changed records identified by the change data feed.
LE
Lettie
1 year ago
B) Convert the batch job to a Structured Streaming job using the complete output mode; configure a Structured Streaming job to read from the customer_churn_params table and incrementally predict against the churn model.
CE
Celeste
1 year ago
A) Apply the churn model to all rows in the customer_churn_params table, but implement logic to perform an upsert into the predictions table that ignores rows where predictions have not changed.
KR
Kristian
1 year ago
Option A is just plain lazy. Why would you want to run the model on every single row when you can be more efficient? That's just asking for trouble.
CA
Caprice
1 year ago
E) Replace the current overwrite logic with a merge statement to modify only those records that have changed; write logic to make predictions on the changed records identified by the change data feed.
AR
Arlene
1 year ago
C) Calculate the difference between the previous model predictions and the current customer_churn_params on a key identifying unique customers before making new predictions; only make predictions on those customers not in the previous predictions.
AL
Alesia
1 year ago
A) Apply the churn model to all rows in the customer_churn_params table, but implement logic to perform an upsert into the predictions table that ignores rows where predictions have not changed.
RE
Renea
1 year ago
I'm torn between Options C and D. Either one could work, but D seems a bit more straightforward with the timestamp field.
JI
Jin
1 year ago
Yes, using the current timestamp field for identification could simplify the process of making predictions on updated records.
TA
Tammy
1 year ago
Option D with the timestamp field does seem straightforward and reliable for identifying changed records.
LA
Laquita
1 year ago
I agree, calculating the difference between previous predictions and current data could be effective.
JO
Johna
1 year ago
Option C seems like a good approach to identify changed records efficiently.
RO
Rory
2 years ago
Option E is my pick. Merging the changed records and making predictions on those is a clean and maintainable solution.
VA
Val
1 year ago
I think option C could also work well by calculating the difference before making new predictions.
ST
Stevie
1 year ago
I agree, option E seems like the most efficient way to handle the changed records.
AR
Ardella
2 years ago
I think Option B is the way to go. Using Structured Streaming to incrementally process the data and make predictions is a more scalable and robust approach.
JA
Jarvis
1 year ago
E) Replace the current overwrite logic with a merge statement to modify only those records that have changed; write logic to make predictions on the changed records identified by the change data feed.
RI
Rikki
1 year ago
C) Calculate the difference between the previous model predictions and the current customer_churn_params on a key identifying unique customers before making new predictions; only make predictions on those customers not in the previous predictions.
WI
Wilbert
1 year ago
B) Convert the batch job to a Structured Streaming job using the complete output mode; configure a Structured Streaming job to read from the customer_churn_params table and incrementally predict against the churn model.
FL
Florinda
2 years ago
I see both points, but I think option B could also be a good approach to consider.
GE
Gerald
2 years ago
Option C makes the most sense to me. Calculating the difference between previous and current predictions is the most efficient way to identify the changes and only run the model on those.
DW
Dwight
2 years ago
I disagree, I believe option E would be more efficient in this scenario.
JA
Janine
2 years ago
I think option C would simplify the identification of changed records.
IV
Ivory
2 years ago
I bet the data engineer is wishing they had a magic wand to deal with all those nested fields. *waves wand*
LO
Lottie
2 years ago
C) Calculate the difference between the previous model predictions and the current customer_churn_params on a key identifying unique customers before making new predictions; only make predictions on those customers not in the previous predictions.
BA
Barbra
2 years ago
B) Convert the batch job to a Structured Streaming job using the complete output mode; configure a Structured Streaming job to read from the customer_churn_params table and incrementally predict against the churn model.
TE
Terry
2 years ago
A) Apply the churn model to all rows in the customer_churn_params table, but implement logic to perform an upsert into the predictions table that ignores rows where predictions have not changed.