Sake

• A modular machine learning framework for budget execution & data analysis built in Python with Scikit, XGBoost, PyTorch, and TensorFlow. Designed for rapid experimentation, visualization, and benchmarking of both classification and regression models, it provides a structured yet extensible workflow that’s equally useful for teaching, prototyping, and real-world application development.

🔬 Data Source

• File A (Account Balances) published monthly by agencies on USASpending
• Required by the DATA Act.
• Pulled automatically from data in the Governmentwide Treasury Account Symbol Adjusted Trial Balance System (GTAS)
• Contains Budgetary resources, obligation, and outlay data for all the relevant Treasury Account Symbols (TAS) in a reporting agency.
• It includes both award and non-award spending (grouped together), and crosswalks with the SF 133 report.

🔄 Unified Evaluation Pipeline

A single interface train_and_evaluate() to:

• Train models
• Cross-validate with nested k-fold
• Generate predictions
• Output evaluation plots & performance metrics
• Store results & timings for meta-analysis

🧪 How to Run

git clone https://github.com/your-username/balance-projector.git
cd balance-projector
pip install -r requirements.txt
jupyter notebook balances.ipynb

🎯 Quickstart

Option A — Google Colab (no local setup)

1. Click the **Open In Colab** badge above.
2. Upload your CSV or mount Google Drive.
3. Set `DATA_PATH` near the top of the notebook.
4. **Runtime → Run all**.

Option B — Local (conda or venv)

bash
# 1) Create environment
conda create -n sake python=3.11 -y
conda activate sake

# 2) Install dependencies
pip install -U pip wheel setuptools
pip install pandas numpy scipy matplotlib seaborn scikit-learn jupyter

# 3) Launch Jupyter
jupyter notebook

Open ipynb/schedule-x.ipynb and run cells top-to-bottom.

📊 Rich Visualization Toolkit

• Confusion Matrix Heatmaps 🔥
• ROC & Precision-Recall Curves 📈
• Actual vs. Predicted Scatterplots 🎯
• Residual Analysis & Error Distribution 🎭
• Feature Importance Charts 📊

⏱️ Timing & Benchmarking

• Automatically logs fit and predict durations
• Model performance rankings across tasks
• Output available in tabular format for export

💡 Custom Dataset Support

• Accepts CSVs, Excel files, or Pandas DataFrames
• Label encoding, numeric coercion, missing data handling
• Drop-in replacement for datasets via parameter injection

🧪 Research Ready

• Benchmark dozens of models easily
• Plug-in architecture for testing experimental models
• Use in classrooms to demo interpretability, overfitting, and variance

📊 Descriptive Statistics

Statistic	Description	Use in Budget Analysis
Mean	Average value	Avg. Outlays, Obligations, etc., across accounts
Median	Middle value	Robust central tendency in skewed financial data
Mode	Most frequent value	Identify common MainAccountCodes or Availability categories
Standard Deviation	Spread around the mean	Indicates variability in execution rates or balances
Variance	Square of standard deviation	Used in statistical tests and model diagnostics
Range	Difference between max and min	Measures total spread of financial metrics
Interquartile Range (IQR)	Spread of middle 50% of data	Identifies budget outliers and extreme accounts
Skewness	Asymmetry of distribution	Skewed obligations suggest few accounts dominate totals
Kurtosis	"Peakedness" of distribution	High values indicate outlier-prone financial data

🔍 Inferrential Statistics

Metric	Description	Use in Budget Analysis
Pearson Correlation	Linear relationship between variables	E.g., TotalResources vs. Obligations
Spearman Correlation	Monotonic (rank-based) relationship	More robust to non-linear trends in financial execution
t-test	Compare means between 2 groups	Discretionary vs. Mandatory accounts' execution rates
ANOVA	Compare means across multiple groups	Obligations across availability periods or account types
Chi-square Test	Categorical independence	Are Main Account Codes related to availability or a specific agency?
Confidence Intervals	Estimate range of a population mean	Upper and lower bound expected obligations or recoveries
Regression Coefficients (p-values)	Test variable significance	Are Recoveries a significant predictor of UnobligatedBalance?
F-statistic (overall regression)	Test whole model fit	Determines the combined influence of all predictors
Z-score / Outlier Tests	Deviation from standard mean	Identify abnormal balances or lapse rates
Boxplots	Visual outlier detection	Discover obligation anomalies within agencies

✅ Classification:

Model	Module
Logistic Regression	sklearn.linear_model.LogisticRegression
SVM	sklearn.svm.SVC
Decision Tree	sklearn.tree.DecisionTreeClassifier
Random Forest	sklearn.ensemble.RandomForestClassifier
XGBoost Classifier	xgboost.XGBClassifier
K-Nearest Neighbors	sklearn.neighbors.KNeighborsClassifier
Gaussian Naive Bayes	sklearn.naive_bayes.GaussianNB
Extra Trees	sklearn.ensemble.ExtraTreesClassifier
Bagging	sklearn.ensemble.BaggingClassifier
AdaBoost	sklearn.ensemble.AdaBoostClassifier

📉 Regression:

Model	Module
Linear Regression	sklearn.linear_model.LinearRegression
Ridge Regression	sklearn.linear_model.Ridge
Lasso Regression	sklearn.linear_model.Lasso
ElasticNet	sklearn.linear_model.ElasticNet
Support Vector Regressor	sklearn.svm.SVR
Decision Tree Regressor	sklearn.tree.DecisionTreeRegressor
Random Forest Regressor	sklearn.ensemble.RandomForestRegressor
Gradient Boosting Regressor	sklearn.ensemble.GradientBoostingRegressor
XGBoost Regressor	xgboost.XGBRegressor
K-Nearest Neighbors	sklearn.neighbors.KNeighborsRegressor
AdaBoost Regressor	sklearn.ensemble.AdaBoostRegressor
Extra Trees Regressor	sklearn.ensemble.ExtraTreesRegressor

📦 Dependencies

Package	Description	Link
numpy	Numerical computing library	numpy.org
pandas	Data manipulation and DataFrames	pandas.pydata.org
matplotlib	Plotting and visualization	matplotlib.org
seaborn	Statistical data visualization	seaborn.pydata.org
scikit-learn	ML modeling and metrics	scikit-learn.org
xgboost	Gradient boosting framework (optional)	xgboost.readthedocs.io
torch	PyTorch deep learning library	pytorch.org
tensorflow	End-to-end ML platform	tensorflow.org
openai	OpenAI’s Python API client	openai-python
requests	HTTP requests for API and web access	requests.readthedocs.io
PySimpleGUI	GUI framework for desktop apps	pysimplegui.readthedocs.io
typing	Type hinting standard library	typing Docs
pyodbc	ODBC database connector	pyodbc GitHub
fitz	PDF document parser via PyMuPDF	pymupdf
pillow	Image processing library	python-pillow.org
openpyxl	Excel file processing	openpyxl Docs
soundfile	Read/write sound file formats	pysoundfile
sounddevice	Audio I/O interface	sounddevice Docs
loguru	Structured, elegant logging	loguru GitHub
statsmodels	Statistical tests and regression diagnostics	statsmodels.org
dotenv	Load environment variables from .env	python-dotenv GitHub
python-dotenv	Same as above (modern usage)	python-dotenv

📁 Customize Dataset

Replace dataset ingestion cell with:

import pandas as pd
df = pd.read_csv("your_dataset.csv")
X = df.drop("target_column", axis=1)
y = df["target_column"]

📊 Outputs

• R², MAE, MSE for each model
• Bar plots of performance scores
• Visual predicted vs. actual scatter charts
• Residual error analysis

Disclaimer: This is for analytical exploration, research, and education purposes.
This is not an official government product; validate against authoritative sources before use.

📝 License

Sake is published under the MIT General Public License v3.

---