Installation

Mon, 01 Jan 0001 00:00:00 +0000

Installation #

Step 1. Clone Repository and cd #

HTTPS

git clone https://github.com/yindrew/DeepRiverDivers.git
cd DeepRiverDivers

SSH

git clone git@github.com:yindrew/DeepRiverDivers.git
cd DeepRiverDivers

GitHub CLI

gh repo clone yindrew/DeepRiverDivers
cd DeepRiverDivers

Step 2. (Optional) Intialize Virtual Environment #

Native Python venv

python -m venv .venv
source .venv/activate.fish # depends on your shell and OS

Using uv

uv venv .venv # replace .venv by other names for virtual env names
source .venv/activate.fish # depends on your shell and OS

Step 3. Install python packages #

Native pip

pip3 install -r requirements.txt

Using uv

uv pip install -r requirements.txt

Commands

Mon, 01 Jan 0001 00:00:00 +0000

Commands #

For what follows, it is assumed that the commands are ran from the root of the project.

`trainer.py` #

This script is used to run trainer. The command:

python3 trainer.py

runs the trainer with default configs. The default configs are given in configs/default.json; currently, it should be the consistent with schemas/model_config.json.

trainer.py can take in an additional parameter for running specified config file:

python3 trainer.py --config my_config.json

which would read in configs/my_config.json, and update the default parameters (that is: the unspecified parameters will use the default ones).

Configs

Mon, 01 Jan 0001 00:00:00 +0000

Configs #

Config parameters configures training process and model architectures, as well as routine i/o setups.

Full Parameters #

Folded

{
 "$schema": "../schemas/model_config_schema.json",
 "general": {...},
 "training_process": {...},
 "action_encoder": {...},
 "card_encoder": {...},
 "cross_attention": {...},
 "output_mlp": {...}
}

Full Defaults

{
 "$schema": "../schemas/model_config_schema.json",
 "general": {
 "device": "cpu",
 "seed": 42069,
 "load_checkpoint_name": "default",
 "save_checkpoint_name": "default",
 "type_of_checkpoint": "final"
 },
 "training_process": {
 "batch_size": 32,
 "num_epochs": 300,
 "learning_rate": 0.0005,
 "weight_decay": 0.01,
 "optimizer": "AdamW",
 "momentum": 0,
 "dataset": "GTO",
 "warm_start": false,
 "p_train_test_split": 0.2,
 "min_epochs": 100,
 "patience": 50
 },
 "action_encoder": {
 "d_model": 32,
 "nhead": 4,
 "dim_feedforward": 128,
 "dropout": 0.1,
 "num_actions": 5,
 "num_positions": 6,
 "num_streets": 5,
 "action_embedding_dim": 4,
 "position_embedding_dim": 4,
 "street_embedding_dim": 4
 },
 "card_encoder": {
 "d_model": 32,
 "nhead": 4,
 "dim_feedforward": 128,
 "dropout": 0.1,
 "num_ranks": 13,
 "num_suits": 4,
 "num_streets": 4,
 "rank_embedding_dim": 8,
 "suit_embedding_dim": 4,
 "street_embedding_dim": 4
 },
 "cross_attention": {
 "num_heads": 4,
 "d_model": 32
 },
 "output_mlp": {
 "d_model": 32
 }
}

Schema

{
 "$schema": "http://json-schema.org/draft-07/schema#",
 "title": "ModelConfig",
 "type": "object",
 "properties": {
 "general": {
 "type": "object",
 "properties": {
 "device": { "type": "string", "enum": ["cpu", "cuda"] },
 "seed": { "type": "integer" },
 "load_checkpoint_name": { "type": "string" },
 "save_checkpoint_name": { "type": "string" },
 "type_of_checkpoint": { "type": "string", "enum": ["best", "final"] }
 }
 },
 "training_process": {
 "type": "object",
 "properties": {
 "batch_size": { "type": "integer" },
 "num_epochs": { "type": "integer" },
 "learning_rate": { "type": "number" },
 "weight_decay": { "type": "number" },
 "optimizer": { "type": "string", "enum": ["Adam", "SGD", "AdamW"] },
 "momentum": { "type": "number" },
 "dataset": { "type": "string", "enum": ["GTO", "Human"] },
 "warm_start": { "type": "boolean" },
 "p_train_test_split": { "type": "number" },
 "min_epochs": { "type": "integer" },
 "patience": { "type": "integer" }
 }
 },
 "action_encoder": {
 "type": "object",
 "properties": {
 "d_model": { "type": "integer" },
 "nhead": { "type": "integer" },
 "dim_feedforward": { "type": "integer" },
 "dropout": { "type": "number" },
 "num_actions": { "type": "integer" },
 "num_positions": { "type": "integer" },
 "num_streets": { "type": "integer" },
 "action_embedding_dim": { "type": "integer" },
 "position_embedding_dim": { "type": "integer" },
 "street_embedding_dim": { "type": "integer" }
 }
 },
 "card_encoder": {
 "type": "object",
 "properties": {
 "d_model": { "type": "integer" },
 "nhead": { "type": "integer" },
 "dim_feedforward": { "type": "integer" },
 "dropout": { "type": "number" },
 "num_ranks": { "type": "integer" },
 "num_suits": { "type": "integer" },
 "num_streets": { "type": "integer" },
 "rank_embedding_dim": { "type": "integer" },
 "suit_embedding_dim": { "type": "integer" },
 "street_embedding_dim": { "type": "integer" }
 }
 },
 "cross_attention": {
 "type": "object",
 "properties": {
 "num_heads": { "type": "integer" },
 "d_model": { "type": "integer" }
 }
 },
 "output_mlp": {
 "type": "object",
 "properties": {
 "d_model": { "type": "integer" }
 }
 }
 }
}

Detailed Descriptions #

At the top level, there are seven accepted keys:

Running Our Codes on DeepRiverDivers Doc

Installation

Installation #

Step 1. Clone Repository and cd #

Step 2. (Optional) Intialize Virtual Environment #

Step 3. Install python packages #

Commands

Commands #

trainer.py #

Configs

Configs #

Full Parameters #

Detailed Descriptions #

`trainer.py` #