Index

Symbols

1D convolution

2012 leap second case study

A

a class missing from the training data failure mode accumulation mode accuracy metric activation function Adam, [1] AdamW agent

algorithm dynamics an input that never moved failure mode anomaly detection Apache Arrow ARX autocorrelation automatic differentiation, [1] forward mode reverse mode

B

B-tree backpressure baseline batch normalization batch pipeline batch processing, [1] bias correction

bias-variance trade-off big data hubris Boeing 737 MAX case study broadcast target shape failure mode broadcasting

C

caching case study 2012 leap second Boeing 737 MAX Google Flu Trends LTSF-Linear M4 forecasting competition Meese and Rogoff exchange-rate forecasts Public Health England case loss reversed causality under closed-loop control turbofan RUL 1D-CNN channel child run classification clock drift failure mode

closed-loop identification column projection column store columnar file compression computation graph confusion matrix container content hash context window continuous-time model convolution convolutional network COPY cross-entropy cross-validation

D

data contract data leakage, [1] data validation, [1] data versioning database index dataframe DataLoader decision boundary decision threshold decision tree delivery semantics dependency resolution design matrix

detection delay deviation variable device placement direct forecasting discrete-time model document store double precision dropout dtype DuckDB DuckDB as a shared server failure mode DVC dynamic process

E

eager execution early stopping embedded database, [1] event time

experiment tracking, [1] EXPLAIN ANALYZE extrapolating outside the training data failure mode

F

failure mode a class missing from the training data an input that never moved broadcast target shape clock drift DuckDB as a shared server extrapolating outside the training data float32 precision loss forecasting across a regime change forgetting zero_grad NULL in a comparison point writes to a column store prompt injection recursive error accumulation reproducible but wrong shifting across a run boundary shuffling a time series before splitting splitting one group across folds undocumented fill value unscaled inputs

false-alarm rate FastAPI feature engineering feature table fine-tuning first-order process float32 float32 precision loss failure mode float64 forecast horizon forecasting across a regime change failure mode foreign key forgetting zero_grad failure mode forward chaining forward fill

G

Gaussian process regression Gini impurity Google Flu Trends case study GPU

gradient accumulation gradient clipping Great Expectations grid search GroupKFold, [1]

H

hidden technical debt

hyperparameter hypertable

I

idempotency

identifiability integration failure

J

JAX

jaxpr jit

K

Kafka

kernel key-value store

L

L-BFGS lag plot lasso late data lazy evaluation learning curve learning rate learning-rate schedule least squares

lockfile log logistic regression long format look-ahead bias loss function, [1] LSTM LTSF-Linear case study

M

M4 forecasting competition case study marginal likelihood MASE metric Meese and Rogoff exchange-rate forecasts case study message passing metric accuracy MASE precision recall RMSE

micro-batch mixed precision MLflow MLflow run MLOps model capacity model registry momentum multilayer perceptron

N

naive forecast see persistence forecast NARX neural network no free lunch theorem

no_grad normalization notebook to module NULL in a comparison failure mode numeric

O

offset OLAP OLTP

optimizer Optuna overfitting

P

padding pandera parameter sharing parent run Parquet partitioning persistence forecast persistent excitation point writes to a column store failure mode Polars pooling positional encoding PostgreSQL precision metric

predicate pushdown primary key PRNG key processing time project scaffold prompt injection failure mode prompting provenance, [1] Public Health England case loss case study pure function pyproject.toml PyTorch

Q

quarantine

R

random search random seed, [1] random walk recall metric receptive field recurrent network recurrent neural network recursive error accumulation failure mode recursive forecasting regression vector regularization relational database

relational model reproducible but wrong failure mode resampling resolver retrieval-augmented generation reversed causality under closed-loop control case study RMSE metric rolling origin rolling window row group row store run boundary

S

sampling interval schema check self-attention sensor calibration sensor drift session window shifting across a run boundary failure mode shuffling a time series before splitting failure mode skill score sliding window

softmax splitting one group across folds failure mode staging table state-space model stationarity steady state steady-state gain stochastic gradient descent stream processing streaming stride supervised learning

T

target leakage temporal convolutional network Tennessee Eastman process tensor test set three-valued logic time constant time plot time-series database TimeSeriesSplit

timestamptz token TPE see Tree-structured Parzen Estimator training loop training set Tree-structured Parzen Estimator trigger tumbling window turbofan RUL 1D-CNN case study

U

undocumented fill value failure mode unit root test

unscaled inputs failure mode unsupervised learning uv

V

validation curve validation set vector database

vectorization version control virtual environment vmap

W

watermark weight decay, [1]

white noise window function windowing

X

XLA

Z

zero-order hold