Fintuna: Parameter optimization for finance
Fintuna is a framework that uses machine learning for asset management. It enables fast prototyping for multi-asset applications such as stock-picking.
- Features:
model training
hyper-parameters tuning
walk-forward backtesting
strategy evaluation
It is a lightweight framework that combines LightGBM, Optuna, Quantstats and Shap to develop ML-based stock-picking strategies.
Multi Asset
Looking at multiple assets is supposed to reveal more alpha-opportunities than looking at a single one. Also, the more assets the more data which is beneficial for machine learning tasks. Therefore, Fintuna is designed for multi-asset applications. The data structure is a Pandas Multiindex Dataframe where the index is time, the first column-level is the asset and the second is the feature (= panel or longitudinal data). Internally features are stacked and a model is trained to learn cross-asset patterns.
# |
Asset 1 |
Asset 2 |
Asset 3 |
Asset 4 |
||||
|---|---|---|---|---|---|---|---|---|
# |
feature1 |
feature2 |
feature1 |
feature2 |
feature1 |
feature2 |
feature1 |
feature2 |
t0 |
float |
category |
float |
category |
float |
category |
float |
NaN |
t1 |
float |
category |
float |
category |
float |
category |
float |
NaN |
Strategy Agnostic
Fintuna is not tied to one specific trading strategy. Strategies are implemented as fintuna.model.ModelBase. It defines the classification task (= extract_label) as well as a a classification-to-returns mapping (= realized_returns). A simple example is to predict the directional change and buy the asset with the most confident prediction (see fintuna.model.LongOnly).
Backtesting
Fintuna uses walk-forward backtesting.
Train data is used to train the classifier.
Tune data is used for hyper-parameter optimization.
Eval data is used for backtesting
Executing the fintuna.Finstudy.explore method multiple times on same data introduces the risk of overfitting. Use feature importance and shap values, rather than merely looking at trading performance.
Calling fintuna.Finstudy.finish prepares the model for deployment. It sub-selects models that also perform well on evaluation data. and refits them on all data.
Data First
A good trading strategy demands good and possibly unique data. Fintuna does NOT help you in finding the right data. But consider the following guidelines:
Have at least a few hundreds of observations.
Use multiple assets.
Use assets with similar characteristics (e.g. cryptos, tech-stocks, etc.).
Make sure features across assets have similar properties (otherwise use zscore).
Use lagged features to boost performance.
Usage
Install fintuna via pip.
pip install fintunaRun the most basic example below. For detailed guidance look at examples at docs or at docs/source/examples.
import fintuna as ft
# get data
data, specs = ft.data.get_crypto_features()
# explore
crypto_study = ft.FinStudy(ft.model.LongOnly, data, data_specs=specs)
results = crypto_study.explore(n_trials=50, ensemble_size=3)
# analyze
ft.utils.plot_backtest(results['performance'], results['benchmark'])TODO
Binance Trading Sink
MajorityVoteEnsemble
Backtest plots with shap values
1 maintainer
Metadata
1 maintainer
Release files for fintuna 0.1.5
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| fintuna-0.1.5.tar.gz | 36.9 MB | Without provenance | Details for fintuna-0.1.5.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| fintuna-0.1.5-py3-none-any.whl | Python 3 | none | any | Details for fintuna-0.1.5-py3-none-any.whl |
Total release size: 73.7 MB
Release files / fintuna-0.1.5.tar.gz
| Download URL | fintuna-0.1.5.tar.gz |
|---|---|
| Size | 36.9 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
7527aeb6a792490e4dfd135aa387dab65911b00337de731ad015e8ef3a1a3ebc
|
|
BLAKE2b-256 checksum How to use checksums |
82b4db4270ba2c4e333eef01712ada1114838815566cd04d1e58a16b55458152
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.1 CPython/3.8.13
|
Release files / fintuna-0.1.5-py3-none-any.whl
| Download URL | fintuna-0.1.5-py3-none-any.whl |
|---|---|
| Size | 36.9 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
d0b9c2b3987d5d33170020b5f6819d555f196595ef2574a54df3973e7ec2dd65
|
|
BLAKE2b-256 checksum How to use checksums |
ae478558db24d5aa8df324156e7cf1c12b60040194707dee108d184cde5e52c5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.1 CPython/3.8.13
|
Security for fintuna 0.1.5
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
