Skip to main content

> ML_ALGORITHMS_ATLAS_v1.0

Algorithms & Method Families

100 qualified algorithmic method families across 9 disciplines: Classical Supervised, Unsupervised, Time Series, Anomaly Detection, Deep Learning, Recommenders, Causal Inference, Reinforcement Learning, and Specialized Methods.

Accelerator Matrix14D Decision Engine
Showing 10 of 100 Method Families (Page 6 of 6)Mathematically Formulated
Algorithm
O(1) recursiontime-series-forecasting

Holt-Winters Exponential Smoothing (ETS)

Level, trend, and seasonal components can be plotted directly and independently inspected.

Implementing Tools:
statsmodelsCaret r
time series forecastinginventory demand planning
#exponential-smoothing-holt-wintersDetails
Algorithm
O(horizon)time-series-forecasting

Prophet (Additive Decomposable Time Series)

Provides clear decomposed component plots for trend, weekly seasonality, annual seasonality, and specific holidays.

Implementing Tools:
ProphetStan
business metric forecastingcapacity planning
Paper (2018)Details
Algorithm
O(samples * horizon * LSTM_pass)time-series-forecasting

DeepAR (Autoregressive Recurrent Networks)

Outputs full Monte Carlo sample paths for probabilistic prediction intervals (p10, p50, p90).

Implementing Tools:
GluonTSPytorch forecasting
large scale demand forecastingprobabilistic forecasting
Paper (2020)Details
Algorithm
O(seq_len^2 * d)time-series-forecasting

Temporal Fusion Transformer (TFT)

Variable Selection Networks yield explicit feature importance weights for static, past, and future inputs, while attention heads reveal temporal dynamics.

Implementing Tools:
Pytorch forecastingGluonTS
multi horizon forecastingcomplex exogenous forecasting
#temporal-fusion-transformer-tftDetails
Algorithm
O(blocks * MLP)time-series-forecasting

N-BEATS (Neural Basis Expansion Analysis)

Interpretable configuration restricts output expansion coefficients to explicit monotonic polynomial trend and periodic harmonics.

Implementing Tools:
NeuralForecastDartsPytorch forecasting
univariate time series forecastingzero shot ts forecasting
#neural-basis-expansion-n-beatsDetails
Algorithm
O(downsampled_MLP)time-series-forecasting

N-HiTS (Neural Hierarchical Interpolation)

Multi-rate pooling enforces that slow stacks synthesize low-frequency macro trend while fast stacks capture high-frequency ripples.

Implementing Tools:
NeuralForecastDarts
long horizon forecastinghigh frequency telemetry
#neural-hierarchical-interpolation-n-hitsDetails
Algorithm
O(channels * (L/P)^2 * d)time-series-forecasting

PatchTST (Channel-Independent Patch Time Series Transformer)

Channel independence prevents spurious cross-variate correlations while patch attention preserves temporal memory.

Implementing Tools:
NeuralForecastTransformers
long horizon forecastingmultivariate time series forecasting
#patchtst-channel-independent-patchingDetails
Algorithm
O(L * log(L) * d)time-series-forecasting

Informer (ProbSparse Attention)

Reduces classical quadratic Transformer memory to O(L log L) via selective active queries.

Implementing Tools:
GluonTSNeuralForecast
long horizon forecastingmultivariate time series forecasting
#informer-probsparse-attentionDetails
Algorithm
O((k * p)^2)time-series-forecasting

Vector Autoregression (VAR & VECM)

Directly yields Granger causality test statistics and orthogonalized impulse response functions (IRF).

Implementing Tools:
statsmodelsCaret r
multivariate econometricsgranger causality analysisimpulse response analysis
#vector-autoregression-varDetails
Algorithm
O(1) memory and time per steptime-series-forecasting

Structured State Space Models (S4 & Mamba)

Hardware-aware selective scan maintains constant memory regardless of sequence length, enabling 1M+ step context windows.

Implementing Tools:
Mamba (mamba-ssm)TransformersPyTorch
long horizon forecastinglong context sequence modelingsignal processing
#structured-state-space-sequence-models-mambaDetails
Showing 91–100 of 100 Method Families