Cloudera Machine Learning Engineer case studies
Cloudera Machine Learning Engineer · Put your knowledge to work in context.
The Cloudera Certified Associate (CCA) Machine Learning Engineer credential is intended for data scientists and ML engineers who build, deploy, and operate models on Cloudera's data and AI platform. This independent study page is designed for candidates searching for the exam: it organizes practice around five practical areas - workbench and project setup, data access and preparation, model development and experimentation, model deployment and serving in Cloudera AI, and the operations, monitoring, and governance of ML jobs. The Cloudera Machine Learning Engineer case studies are 5 cases and about 5,238 words.
1One project, three incompatible runtimes
A seven-person data science team shares a single CML project in a CDP Private Cloud Base workspace. The project now carries three workloads: exploratory notebooks that need Python 3.11 with PyTorch 2.x and CUDA 12, a legacy scoring script pinned to Python 3.8 and scikit-learn 0.24, and a nightly job that builds a feature table.
1,013 words, about 5 min2The table that returns zero rows in the session
A data scientist is building a churn model on risk.events , a Hive table with roughly 1.4 billion rows, partitioned by dt (five years, about 1,800 partitions) and region , and stored as roughly 220,000 small Parquet files. In a CML session they connect with impyla through a project data connection.
1,054 words, about 5 min3The 0.994 AUC that did not survive deployment
An anti-fraud model for card transactions is being built in a CML project. The labelled dataset holds 18 months of history with a 0.6% positive rate.
1,018 words, about 5 min4p99 of 1.2 seconds against a 150 ms contract
A churn model is deployed as a CML model endpoint with three replicas of 4 vCPU and 16 GB. The contract is p99 ≤ 150 ms at 40 requests per second, with bursts to 120.
1,057 words, about 5 min5Four nights of successful failures
A nightly CML job starts at 02:00, trains a demand model, and appends scores to ml.scores daily . Median runtime is 50 minutes and a dashboard must show fresh data by 06:00.
1,096 words, about 5 min