Machine learningProject record · Sep 2026

Machine Learning & Model Research

A-share representation and attention-model adaptations

Chart-image CNN/ViT, market-guided attention, and cross-sectional representation studies in one model-research family.

PyTorchCNNVision TransformerMASTERt-SNEA-share dataAttention factorsStatistical arbitragePoint-in-time panels
Why it sits here. Placed by the scope of implementation, the available contribution evidence, and the distinct technical capability it demonstrates.
03 / Somewhere between data and understanding.STUDY IN SPACE

01 / IMPLEMENTATION & CONTRIBUTION

What the work involves

Local implementations and adaptations explore chart-image CNNs/ViTs, MASTER adapters, and Attention Factors. Each original paper and model remains credited to its authors.

Technical depth

Reproducible chart rendering, visual-versus-numeric comparisons, market-conditioned adapters, characteristic attention, variable-universe masks, PIT alignment and portfolio construction under market constraints.

The project family

A_Share/factorResearch/CNNA_Share/factorResearch/vit-sdf-lassoA_Share/factorResearch/MASTERA_Share/factorResearch/visualizationA_Share/attention_factors

02 / RESULTS

What came out of it

The family includes chart rendering, model and evaluation modules, and ViT inference/visualization. Attention Factors distinguishes the paper’s long/short reference from a long-tilt hedge construction.

03 / SUPPORTING EVIDENCE

Follow the source

Implementation notes, project records, and supporting artifacts.

Source context & project scope

No paper publication, successful SDF replication, model advantage or deployment is established. Planned visual fusion and directly executable market variants must not be conflated with finished evaluation.

Chart rendering, CNN baseline, feature controls, fusion and evaluation research layout.

SOURCE · 2026-09-17

Explicit A-share adaptation and active workflow boundaries.

SOURCE · 2026-09-17

Local adaptation source exists on the upstream MASTER checkout.

SOURCE · 2026-09-17

Local dataset adapter accompanies feature and market-input modules.

SOURCE · 2026-09-17

Upstream attribution and separate reference/deployable-construction designs.

SOURCE · 2026-09-17

Training module exists with model, cost and backtest code.

SOURCE · 2026-09-17

Alternative portfolio construction source.

SOURCE · 2026-09-17
CONTINUE IN MACHINE LEARNING & MODEL RESEARCH

Leader-follower rotation and event ranking