maxzhangalt/sgai-team-alien

Team Alien Deluxe RL Dataset

Dataset for Team Alien's Deluxe RL data collection and analysis for Ameen later.

Repository Structure

├── base/
│   ├── baselineRL_harish/
│   │   ├── episodes.csv
│   │   ├── steps.csv
│   │   └── eval.md
│   ├── baselineRL_karunya/
│   │   ├── episodes.csv
│   │   ├── steps.csv
│   │   └── eval.md
│   └── models/
│       ├── harish.pth
│       └── karunya.pth
└── deluxe/

Base Game Mode

The base/ directory contains datasets, evaluation results, and trained models for the base game mode.

Baseline Directories

The following directories contain results generated by the corresponding baseline agents:

• base/baselineRL_harish/
• base/baselineRL_karunya/

Each baseline directory contains the files described below.

episodes.csv

Contains episode-level statistics for each complete game run.

• Approximate number of rows: 1,000
• Episode identifier: run_id
• Columns:

model_id,
run_id,
seed,
end_reason,
n_save,
n_skip,
n_squish,
n_scram,
n_eject,
n_turns,
saved,
killed,
saved_value,
reward

Each row represents one complete game episode.

steps.csv

Contains detailed, step-level information about the actions and decisions made during each game episode.

• Approximate number of rows: 40,000
• Episode identifier: run_id
• Step identifier: step_index
• Columns:

model_id,
run_id,
seed,
step_index,
turn_index,
decision_kind,
action,
chosen_action_index,
q_values,
action_mask,
greedy,
obs_json,
humanoid_fp,
true_state,
is_carrier,
carrier_prob,
displayed_carrier_pct,
remaining_time,
slots_used,
slots_total,
n_passengers,
amb_healthy,
amb_injured,
amb_zombie,
amb_corpse,
amb_carrier_healthy,
amb_carrier_injured,
ambulance_value,
carrier_posterior,
p_safe_scram,
ejected_slot,
ejected_true_state,
ejected_is_carrier,
ejected_carrier_prob,
ejected_displayed_pct

Rows can be grouped by run_id to reconstruct and analyze the complete sequence of decisions within an episode.

eval.md

Contains evaluation summaries and terminal metrics for reviewing each model’s overall behavior and performance at a glance.

Trained Models

The base/models/ directory contains the trained reinforcement learning model checkpoints:

base/models/
├── harish.pth
└── karunya.pth

Deluxe Game Mode

The deluxe/ directory is reserved for datasets, evaluation results, and trained models generated in the Deluxe game mode.

This directory is currently empty.

When Deluxe data is added, update this dataset card.

License

This dataset repository is licensed under the MIT License.