ldt-toolkit 0.0.4

Install the following first:

• Python 3.10 to 3.12 AND OR uv
• Go (for ldt CLI)
• R + Rscript (required only when running LCGA/GMM tools through lcmm)

Quick checks:

python --version || python3 --version # Not necessary if you use `uv`
uv --version
go version
Rscript --version

Example installs by OS for uv go and R:

# macOS (Homebrew)
brew update
brew install uv go r

# Linux (Ubuntu/Debian)
sudo apt-get update
sudo apt-get install -y golang r-base
curl -LsSf https://astral.sh/uv/install.sh | sh

If you plan to run LCGA/GMM, install R-side dependencies:

Rscript --vanilla -e "repos <- 'https://cloud.r-project.org'; required_packages <- c('lcmm'); missing <- setdiff(required_packages, rownames(installed.packages())); if (length(missing)) install.packages(missing, repos = repos) else message('All required R packages are already installed.')"
# 👇 If you have cloned the repo, you can also run:
Rscript --vanilla setup_R/install_requirements.R

from pathlib import Path

from ldt.data_preparation import EventShockRecovery, MissingDataScenarios, TrendPatterns

out = Path("/path/to/your/project/data")
out.mkdir(parents=True, exist_ok=True)

TrendPatterns(n_samples=400, n_waves=5, random_state=7).prepare().to_csv(
    out / "synthetic_trend_patterns.csv", index=False
)

EventShockRecovery().prepare(
    n_samples=400,
    n_waves=5,
    random_state=7,
    feature_cols=["depressive_score"],
    shock_wave=3,
    shock_mean=3.5,
    recovery_rate=0.9,
    noise_sd=0.8,
).to_csv(out / "synthetic_event_shock.csv", index=False)

MissingDataScenarios().prepare(
    n_samples=400,
    n_waves=5,
    random_state=7,
    feature_cols=["depressive_score", "sleep_score"],
    mechanism="mixed",
    missing_rate=0.20,
    dropout_rate=0.10,
    mar_strength=1.0,
).to_csv(out / "synthetic_with_missing.csv", index=False)

from pathlib import Path

from ldt.data_preprocessing import BuildTrajectories, ShowTable

input_long = Path("/path/to/your/project/data/synthetic_trend_patterns.csv")
out = Path("/path/to/your/project/outputs")
out.mkdir(parents=True, exist_ok=True)

BuildTrajectories().fit_preprocess(
    mode="from_scratch",
    input_path=input_long,
    output_path=out / "trajectories_dtw_kmeans.csv",
    id_col="subject_id",
    time_col="wave",
    value_cols=["depressive_score"],
    builder="dtw_kmeans",
    n_trajectories=4,
)
BuildTrajectories().fit_preprocess(
    mode="from_scratch",
    input_path=input_long,
    output_path=out / "trajectories_clusterMLD.csv",
    id_col="subject_id",
    time_col="wave",
    value_cols=["depressive_score"],
    builder="clusterMLD",
    n_trajectories=4,
)

ShowTable().fit_preprocess(
    input_path=out / "trajectories_clusterMLD.csv",
    output_html=out / "trajectories_clusterMLD_report.html",
    open_browser=False,
)

from pathlib import Path

from ldt.data_preparation import MissingDataScenarios
from ldt.data_preprocessing import (
    AggregateLongToCrossSectional,
    BuildTrajectories,
    CleanDataset,
    CombineDatasetWithTrajectories,
    MissingImputation,
)
from ldt.machine_learning import StandardMachineLearning

root = Path("/path/to/your/project")
raw_long = root / "data/synthetic_long_with_missing.csv"
clean_long = root / "outputs/long_clean.csv"
imputed_long = root / "outputs/long_imputed.csv"
trajectories = root / "outputs/trajectories.csv"
cross_sectional = root / "outputs/cross_sectional.csv"
model_ready = root / "outputs/model_ready.csv"

long_df = MissingDataScenarios().prepare(
    n_samples=1200,
    n_waves=6,
    random_state=42,
    feature_cols=["depressive_score", "sleep_score", "anxiety_score"],
    mechanism="mixed",
    missing_rate=0.20,
    dropout_rate=0.15,
    mar_strength=1.10,
)
raw_long.parent.mkdir(parents=True, exist_ok=True)
long_df.to_csv(raw_long, index=False)

CleanDataset().fit_preprocess(input_path=raw_long, output_path=clean_long)
MissingImputation().fit_preprocess(
    technique="median_imputation",
    input_path=clean_long,
    output_path=imputed_long,
)
BuildTrajectories().fit_preprocess(
    mode="from_scratch",
    input_path=imputed_long,
    output_path=trajectories,
    id_col="subject_id",
    time_col="wave",
    value_cols=["depressive_score"],
    builder="clusterMLD",
    n_trajectories=4,
)
AggregateLongToCrossSectional().fit_preprocess(
    input_path=imputed_long,
    output_path=cross_sectional,
    subject_id_col="subject_id",
    numeric_columns=["depressive_score", "sleep_score", "anxiety_score", "age_baseline"],
    numeric_agg="mean",
)
CombineDatasetWithTrajectories().fit_preprocess(
    input_original_data_path=cross_sectional,
    input_trajectories_data_path=trajectories,
    output_path=model_ready,
    original_id_col="subject_id",
    trajectory_id_col="subject_id",
    merge_type="left",
    trajectory_columns=["trajectory_id", "trajectory_name"],
)

ml_result = StandardMachineLearning().fit_predict(
    technique="run_experiment",
    input_path=model_ready,
    target_column="trajectory_id",
    feature_columns="depressive_score,sleep_score,anxiety_score,age_baseline",
    estimator_key="random_forest",
    metric_keys="accuracy,f1_macro",
    cv_folds=5,
    validation_split="none",
    multiclass_mode="multiclass",
    random_seed=42,
    output_dir=str(root / "outputs/standard_ml"),
)
print(ml_result["mean_score"], ml_result["report_path"])