Data Science for Neoclouds
Two sections in one guide. First, the practitioner playbook for ML / DS work on neocloud infrastructure — provider choice, training and fine-tuning workflows, experiment tracking, cost-aware experimentation. Then, the inside view: how DS teams at neocloud companies build pricing models, churn analysis, capacity forecasting, marketplace ranking, and anomaly detection.
Section A · Orient
Section B · DS using neoclouds (practitioner)
01The DS Workflow on GPU CloudsHow ML / DS workflows differ when running on neocloud infrastructure vs hyperscaler default.
02Picking the Right Provider for ML WorkProvider decision framework for the DS / ML perspective — different criteria than DE-side.
03Training & Fine-Tuning WorkflowsEnd-to-end workflow patterns; hyperparameter sweeps; multi-node training; distillation.
04Experiment Tracking & ReproducibilityMLflow / Weights & Biases / Comet / Neptune; tying runs to compute, data, code.
05Cost-Aware ML ExperimentationWhen to use spot, how to allocate compute across experiments, the ML-specific cost tradeoffs.
Section C · DS inside a neocloud
06Pricing ModelsPrice elasticity, segment-based pricing, dynamic pricing on marketplaces, discount optimization.
07Churn & RetentionPredicting churn in GPU-rental customers; saving accounts; reactivation campaigns.
08Capacity ForecastingDemand forecasting, supply planning, the buy-vs-build models that drive cap-ex.
09Marketplace Ranking & MatchingLearning-to-rank for provider listings; matching algorithms; A/B testing on the marketplace.
10Anomaly DetectionFraud detection, reliability anomalies, hardware-failure prediction. The DS side of trust and ops.
Section D · Synthesis