📖 Tier 1: Prepare & Study Guide ✓ 100% Solved with Rationales

Time-Series Forecasting & MLOps Deployment (Artificial Intelligence) Solved Questions & Notes (2026) - Apex Rankers

Artificial Intelligence & Data Science > Artificial Intelligence > Time-Series Forecasting & MLOps Deployment

82 Total Solved Questions
~123 mins Estimated Reading Time
1 Subject Areas / Chapters
Select Topic Area / Chapter: Click any section below to switch questions

Time-Series Forecasting & MLOps Deployment

100%
Showing 25 of 82 (30%)
🎯 Practice
Jump:
Q. 1 Artificial Intelligence
Difficulty: Hard (1 Mark)
In time-series analysis, what three conditions must be satisfied for a series to be considered 'Weakly Stationary' (Covariance Stationary)?
A
1. Constant mean over time, 2. Constant variance over time, 3. Autocovariance depends only on the time lag between observations and not on actual time t
✓ Correct
B
1. Zero mean, 2. Zero variance, 3. Zero correlation
C
1. Infinite trend, 2. Extreme seasonality, 3. Random noise
D
1. Strictly positive numbers, 2. Strictly integer values, 3. Equal length intervals
💡 Step-by-Step Explanation & Concept Rationale
Stationarity ensures statistical properties do not change over time, a prerequisite for linear forecasting models like ARIMA.
Q. 2 Artificial Intelligence
Difficulty: Medium (1 Mark)
What statistical hypothesis test is standardly used to test for the presence of a unit root (non-stationarity) in a univariate time series?
A
Augmented Dickey-Fuller (ADF) Test
✓ Correct
B
Student's t-test
C
Chi-Square test of independence
D
ANOVA F-test
💡 Step-by-Step Explanation & Concept Rationale
The ADF test checks the null hypothesis that a unit root is present (non-stationary); p-value < 0.05 rejects the null, indicating stationarity.
Q. 3 Artificial Intelligence
Difficulty: Medium (1 Mark)
In an ARIMA(p, d, q) model, what do the parameters p, d, and q represent?
A
p = Order of the Autoregressive (AR) terms, d = Degree of differencing (Integration), q = Order of the Moving Average (MA) terms
✓ Correct
B
p = Population, d = Density, q = Quantity
C
p = Precision, d = Dimension, q = Quantile
D
p = Penalty, d = Depth, q = Queue size
💡 Step-by-Step Explanation & Concept Rationale
ARIMA models combine differencing 'd' to achieve stationarity with AR(p) past lags and MA(q) past forecast error terms.
Q. 4 Artificial Intelligence
Difficulty: Hard (1 Mark)
What visual diagnostic plots are examined to determine the appropriate AR order (p) and MA order (q) in Box-Jenkins ARIMA modeling?
A
Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) plots
✓ Correct
B
ROC curve and Precision-Recall curve
C
Confusion matrix and Silhouette plot
D
Scatter plot and Box plot
💡 Step-by-Step Explanation & Concept Rationale
PACF cuts off after lag p for AR(p) processes; ACF cuts off after lag q for MA(q) processes.
Q. 5 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'SARIMA' (Seasonal ARIMA: ARIMA(p,d,q)(P,D,Q)_s) and what does parameter 's' signify?
A
Seasonal ARIMA incorporating seasonal AR, differencing, and MA terms; 's' is the seasonal period length (e.g. s=12 for monthly data, s=24 for hourly)
✓ Correct
B
A model for predicting summer temperatures only
C
A software tool for running ARIMA on GPUs
D
A method that ignores seasonal cycles completely
💡 Step-by-Step Explanation & Concept Rationale
SARIMA captures both non-seasonal short-term dynamics and recurring seasonal cycle patterns (e.g. annual monsoon rainfall peaks).
Q. 6 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'Prophet' (developed by Meta / Facebook) additive time-series forecasting model?
A
A decomposable time-series model: y(t) = g(t) + s(t) + h(t) + epsilon_t, combining non-linear trends with changepoint detection, seasonality (Fourier series), and holiday effects
✓ Correct
B
A religious astrology software
C
A deep learning vision model for satellite photos
D
A recurrent neural network for natural language text
💡 Step-by-Step Explanation & Concept Rationale
Taylor & Letham designed Prophet to handle real-world business and environmental time series with missing data, trend shifts, and robust outlier handling.
Q. 7 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'DeepAR' (Salinas et al. / Amazon Research) for probabilistic time-series forecasting?
A
An autoregressive recurrent neural network (LSTM/GRU) that learns global non-linear representations across thousands of related time series, outputting parametric probability distributions
✓ Correct
B
A model that only forecasts deep ocean water currents
C
A classical linear regression equation
D
A decision tree model for tabular classification
💡 Step-by-Step Explanation & Concept Rationale
DeepAR trains across large item catalogs or multi-sensor networks (e.g. 500 river gauge stations), predicting probabilistic quantile forecasts.
Q. 8 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'Temporal Fusion Transformer' (TFT, Lim et al. / Google Cloud AI)?
A
An attention-based architecture for multi-horizon time-series forecasting that combines Variable Selection Networks, Gated Residual Networks, and Multi-Head Self-Attention with temporal interpretability
✓ Correct
B
A transformer that alters time in video games
C
A high-voltage electrical power transformer
D
A software tool for scheduling calendar meetings
💡 Step-by-Step Explanation & Concept Rationale
TFT handles heterogeneous inputs (static metadata, known future inputs like rain forecasts, and observed past histories) with native prediction intervals.
Q. 9 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'N-BEATS' (Neural Basis Expansion Analysis for Interpretable Time Series, Oreshkin et al.)?
A
A deep neural architecture composed of backward and forward residual link blocks using basis expansions to perform pure deep learning time-series forecasting without recurrent or attention layers
✓ Correct
B
A music rhythm generation software
C
A heartbeat monitoring device
D
A video streaming compression tool
💡 Step-by-Step Explanation & Concept Rationale
N-BEATS surpassed classical statistical ensembles in the M4 competition by decomposing forecasts into interpretable trend and seasonality basis functions.
Q. 10 Artificial Intelligence
Difficulty: Hard (1 Mark)
When using LSTM networks for multi-step river flood discharge forecasting, what is 'Autoregressive Multi-Step' versus 'Direct Multi-Step' forecasting?
A
Autoregressive iteratively feeds single-step predictions back as inputs for subsequent steps; Direct trains separate independent models for each future forecast horizon (t+1, t+2, ..., t+H)
✓ Correct
B
Autoregressive runs on CPU; Direct runs on GPU
C
Autoregressive uses images; Direct uses audio
D
There is no difference in their forecasting architectures
💡 Step-by-Step Explanation & Concept Rationale
Direct forecasting avoids cascading error accumulation inherent in autoregressive rollouts at the cost of training H independent models.
Q. 11 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'MLflow' (open-source MLOps platform) and what are its 4 core components?
A
1. MLflow Tracking (logging experiments, metrics, parameters), 2. MLflow Projects (packaging code), 3. MLflow Models (deploying models), 4. MLflow Model Registry (centralized model lifecycle management)
✓ Correct
B
1. Monitor, 2. Keyboard, 3. Mouse, 4. Processor
C
1. CPU, 2. GPU, 3. RAM, 4. Hard Drive
D
1. Login, 2. Password, 3. Email, 4. Phone
💡 Step-by-Step Explanation & Concept Rationale
MLflow manages the complete machine learning lifecycle from experimentation to model registry and deployment.
Q. 12 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'DVC' (Data Version Control) in modern MLOps pipelines?
A
An open-source Git-extension tool that versions massive datasets, model weights, and intermediate pipeline artifacts using content-addressable storage (S3, GCS, local storage) without bloating Git repositories
✓ Correct
B
A digital video camera recording software
C
A driver verification card for trucks
D
A database management system for SQL tables
💡 Step-by-Step Explanation & Concept Rationale
DVC stores lightweight .dvc pointer files in Git while syncing large multi-gigabyte data files and model checkpoints to remote cloud buckets.
Q. 13 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'Docker' containerization and why is it essential for machine learning reproducibility and deployment?
A
Packaging an ML model along with its code, runtime dependencies, system libraries, CUDA drivers, and Python environment into an isolated, portable container image that runs identically on any infrastructure
✓ Correct
B
A physical shipping container used on cargo vessels
C
A computer monitor calibration tool
D
A file compression utility like WinRAR
💡 Step-by-Step Explanation & Concept Rationale
Containers eliminate 'it works on my machine' dependency conflicts between local development laptops and production cloud clusters.
Q. 14 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'Kubernetes' (K8s) in production machine learning infrastructure?
A
An open-source container orchestration platform that automates the deployment, horizontal scaling, load balancing, health monitoring, and management of containerized ML services across server clusters
✓ Correct
B
A programming language developed by Google
C
A database management tool for NoSQL data
D
A graphical design software for websites
💡 Step-by-Step Explanation & Concept Rationale
K8s manages auto-scaling of inference pods (e.g. via HPA based on GPU utilization or request queues) and zero-downtime rolling updates.
Q. 15 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'KServe' (formerly KFServing) on Kubernetes?
A
A cloud-native model serving framework on Kubernetes providing serverless auto-scaling (including scale-to-zero), canary rollouts, and standardized inference protocol for PyTorch, TensorFlow, ONNX, and XGBoost
✓ Correct
B
A customer service ticketing software
C
A restaurant food ordering application
D
A file transfer protocol for servers
💡 Step-by-Step Explanation & Concept Rationale
KServe provides production ML inference capabilities like payload logging, drift detection integration, and multi-model serving out-of-the-box.
Q. 16 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'ONNX' (Open Neural Network Exchange) runtime format?
A
An open, cross-platform standard representation format for machine learning models that allows models trained in PyTorch/TensorFlow to be executed on optimized C++ inference runtimes across diverse hardware
✓ Correct
B
An online video streaming codec
C
An audio format like MP3
D
A database query language
💡 Step-by-Step Explanation & Concept Rationale
ONNX decouples training frameworks from production serving environments, enabling hardware acceleration via ONNX Runtime on Intel, AMD, ARM, and NVIDIA chips.
Q. 17 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'Triton Inference Server' (developed by NVIDIA)?
A
An open-source, enterprise-grade multi-framework inference serving software that optimizes GPU/CPU utilization through dynamic batching, concurrent model execution, and shared memory pipelines
✓ Correct
B
A gaming console manufactured by NVIDIA
C
A submarine exploration vehicle
D
A graphics design software suite
💡 Step-by-Step Explanation & Concept Rationale
Triton serves PyTorch, TensorRT, ONNX, and Python models simultaneously with microsecond latency and hardware saturation.
Q. 18 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'Dynamic Batching' in Triton Inference Server / vLLM?
A
Automatically grouping independent client inference requests arriving within a small time window (e.g. 2 milliseconds) into a single batch on the GPU to maximize hardware throughput without violating latency SLAs
✓ Correct
B
Batching requests once every 24 hours
C
Combining all training datasets into one file
D
Shuffling data randomly during training
💡 Step-by-Step Explanation & Concept Rationale
Dynamic batching dramatically increases server throughput (QPS) on concurrent traffic while maintaining low response latency for individual users.
Q. 19 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'FastAPI' and why is it preferred over Flask for building production Python ML APIs?
A
FastAPI is built on Starlette and Pydantic, supporting native asynchronous execution (async/await), automatic request data validation, high throughput, and automatic OpenAPI (Swagger) documentation generation
✓ Correct
B
FastAPI only works with fast computers
C
FastAPI requires no programming knowledge
D
FastAPI executes on GPU hardware only
💡 Step-by-Step Explanation & Concept Rationale
FastAPI delivers NodeJS/Go-level performance in Python and simplifies input/output schema validation for complex ML payloads.
Q. 20 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'gRPC' compared to REST (HTTP/JSON) for high-performance ML inference microservices?
A
gRPC uses HTTP/2 with binary Protocol Buffers (Protobuf) serialization, delivering substantially lower serialization overhead, lower network latency, and bidirectional streaming compared to text-based JSON over HTTP/1.1
✓ Correct
B
gRPC is for web browsers only; REST is for servers
C
gRPC requires no internet connection
D
There is no performance difference between them
💡 Step-by-Step Explanation & Concept Rationale
In high-throughput computer vision and audio pipelines, Protobuf serialization in gRPC avoids JSON text parsing bottlenecks.
Q. 21 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'Model Drift' (Model Decay) in production ML systems?
A
The gradual degradation of a model's predictive performance over time caused by evolving real-world conditions, changing user behaviors, or unaligned data distributions
✓ Correct
B
The physical movement of a computer server on a table
C
The loss of electricity to a server room
D
A bug in Python code that deletes files
💡 Step-by-Step Explanation & Concept Rationale
Model drift requires continuous telemetry monitoring and automated retraining triggers to maintain production accuracy.
Q. 22 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'Evidently AI' / 'Whylogs' library used for in MLOps?
A
Continuous monitoring and visualization of data drift, target drift, model quality metrics, and feature distribution shifts in production ML pipelines
✓ Correct
B
Building 3D video games
C
Editing audio podcasts
D
Formatting source code indentation
💡 Step-by-Step Explanation & Concept Rationale
Evidently computes statistical distance tests (KS test, Wasserstein distance, PSI) to generate production drift dashboards.
Q. 23 Artificial Intelligence
Difficulty: Hard (1 Mark)
What is 'Feature Store Online vs Offline Store' architecture?
A
Offline store (Parquet/Snowflake/BigQuery) is optimized for high-throughput batch historical feature retrieval for training; Online store (Redis/DynamoDB) is optimized for ultra-low latency (<10ms) single-record feature lookups for inference
✓ Correct
B
Offline is disconnected from power; Online is connected
C
Offline is for images; Online is for text
D
There is no architectural difference between them
💡 Step-by-Step Explanation & Concept Rationale
Dual-store feature architecture guarantees feature consistency while satisfying opposing throughput and latency requirements.
Q. 24 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'CI/CD for Machine Learning' (CD4ML / MLOps Pipeline)?
A
Continuous Integration, Continuous Delivery, and Continuous Training practices that automate testing of code, data validation, automated model retraining, and zero-downtime deployment
✓ Correct
B
Writing code without testing
C
Deploying models manually by copying files via USB drive
D
Shutting down servers during software updates
💡 Step-by-Step Explanation & Concept Rationale
CD4ML treats data and model pipelines with the same automated rigor as traditional software engineering.
Q. 25 Artificial Intelligence
Difficulty: Medium (1 Mark)
What is 'Great Expectations' tool in data quality engineering?
A
A Python data validation framework that asserts, documents, and monitors data pipeline expectations (e.g. column ranges, nullness, types) before data enters model training
✓ Correct
B
A novel by Charles Dickens
C
A philosophical theory on AI ethics
D
A performance rating given to employees
💡 Step-by-Step Explanation & Concept Rationale
Great Expectations catches corrupt, out-of-bounds, or incomplete data at ingestion checkpoints, preventing silent model training failures.
Study Stream Progress: Showing 25 of 82 Questions (30%)
Jump to:

Ready to Test Your Retention & Speed?

Now that you have reviewed the study questions and rationales, test yourself in our interactive 1-by-1 practice engine or take the full official timed mock exam.