<!-- 
-->
<img src="assets/logo.png" alt="makeprediction logo" width="500px"/>
<!-- <img src="assets/logo_1.png" alt="makeprediction logo" width="300px"/>
-->
MakePrediction is a package for building an automatic Gaussian process regression (GPR) models for time series prediction in Python. It was originally created by [Hanany Tolba].
* MakePrediction is an open source project. If you have relevant skills and are interested in contributing then please do contact us (hananytolba@yahoo.com).*
Gaussian process regression (GPR):
==========================================
The advantages of this Gaussian processes package:
* Very fast training.
* Very fast prediction.
* The prediction can be interpolated as desired.
* The training of the model is automatic:
- *no kernel function needs to be specified*
- *an optimal choice of kernel is automatically elaborated.*
* Possibility to choose a kernel function manually.
* The prediction is probabilistic (Gaussian) so that confidence intervals can be calculated and used to decide whether to make a strategic decision.
* The package provides an API for deployment.
## Where do you find time series?
* Energy
* Finance
* Medical, Biotech, and Healthcare
* IoT Monitoring
* Supply Chain
* Agriculture
* Retail
## What does makeprediction do?
* Modelling and analysis time series.
* Automatic time-series prediction (forecasting).
* Real-Time time series prediction.
* Deploy on production the fitted (or saved) makeprediction model.
### Applications:
* Energy consumption prediction.
* Energy demand prediction.
* Stock price prediction.
* Stock market prediction.
* ...
## Install MakePrediction package
### Latest release from PyPI
* pip install makeprediction
### Latest source from GitHub
*Be aware that the `master` branch may change regularly, and new commits may break your code.*
[MakePrediction GitHub repository](https://github.com/HananyTolba/MakePrediction.git), run:
* pip install .
*or*
* python setup.py install
## Get Started with MakePrediction
### Example
Here is a simple example:
#### Data
```python
import pandas as pd
import numpy as np
import plotly.graph_objects as go
from makeprediction.gpts import GaussianProcessTimeSerie
from makeprediction.kernels import RBF, White, Linear, Periodic, Matern
from makeprediction.visualization import Visualizer
#### generate a random noisy time series
###############################
date = pd.date_range(start = '2022/06',periods = 1000, freq = '20T')
# As sum of some Gaussian kernels
# As sum of some Gaussian kernels
kernel = RBF() + Periodic() + White(variance = .01)
# add mean and variance
data = 100 + 10*kernel.simulate(date, seed = np.random.seed(115))
## generate data without fixe np.random.seed
# data = 100 + 10*kernel.simulate(date)
# create a dataframe with data
df = pd.DataFrame(data = data, index = date, columns=['value'])
print(df.head())
# split time serie into train and test
TRAIN_SIZE = int(.8*len(df))
df_train, df_test = df[:TRAIN_SIZE], df[TRAIN_SIZE:]
# Create an instance of the class GaussianProcessTimeSerie with train data:
#########################################
model = GaussianProcessTimeSerie(df_train.index, df_train.value)
# Show train data with test data
Visualizer.iplot(model, df_test.index, df_test.value)
```
<img src="assets_images/fig0.png" width="900px"/>
#### Train
```python
# fit the model
model.fit()
```
#### Test (long term prediction)
We will first show a simple prediction without updating with new observations (in other words without ever using df_test.value). We can say that the prediction horizon is infinite.
```python
#predict with model and plot result
model.predict(df_test.index)
Visualizer.iplot(model, df_test.index, df_test.value)
```
<img src="assets_images/fig1.png" width="900px"/>
```python
# Model components
#how the components of the model (decomposition)
fig3 = Visualizer.iplot_components(model,return_fig=True)
fig3.write_image("fig3.png",width = 900,height =700, scale = 5)
```
<img src="assets_images/fig3.png" width="900px"/>
```python
#plot only test result
fig2 = Visualizer.iplot(model, df_test.index, df_test.value, return_fig=True, test_only=True)
fig2.write_image("fig2.png",width = 1400,height =700)
fig2.show()
```
<img src="assets_images/fig2.png" width="900px"/>
#### Online prediction with horizon = 1
Contrary to the previous case, the horizon is 1. In other words, we predict a first value at a given time, then we will update the model by communicating this observation via the update method and so on (see codes).
```python
# Online prediction with updating
ypred = np.empty(shape = (0,))
ypred_std = np.empty(shape = (0,))
for x,y in df_test.itertuples():
# predict for x value
yp,yp_std = model.predict(x,return_value = True)
ypred = np.append(ypred,yp)
ypred_std = np.append(ypred_std,yp_std)
# update the model for (x,y)
data = {'x_update': x, 'y_update': y}
model.update(**data)
# Show new prediction
# Show new update prediction
fig2.add_trace(
go.Scatter(
x=df_test.index,
y=ypred,
mode="lines",
name='Prediction(horizon = 1)',
showlegend=True)
)
fig2.write_image("fig2_update_one_head.png",width = 900,height =700, scale = 5)
```
<img src="assets_images/fig2_update_one_head.png" width="900px"/>
#### After updating, how to check?
Note that each time the model is updated with a new data data = {'x_update': x (time), 'y_update': y (value)}
or even data vector. The model learns this data and adjusts itself to become more and more efficient.
```python
#prediction of data already seen(updated)
yupdate, _ = model.predict(df_test.index,return_value = True)
fig2.add_trace(
go.Scatter(
x=df_test.index,
y=ypred,
mode="lines",
name='Prediction(horizon = 1)',
showlegend=True)
)
fig2.add_trace(
go.Scatter(
x=df_test.index,
y=yupdate,
mode="lines",
name='Update model',
showlegend=True)
)
fig2.show()
fig2.write_image("fig2_update.png",width = 900,height =700, scale = 5)
```
<img src="assets_images/fig2_update.png" width="900px"/>
Raw data
{
"_id": null,
"home_page": "https://github.com/HananyTolba/MakePrediction.git",
"name": "makeprediction",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.6",
"maintainer_email": "",
"keywords": "Gaussian Process Regression,Time series prediction,Machine Learning",
"author": "Hanany Tolba",
"author_email": "hananytolba@yahoo.com",
"download_url": "https://files.pythonhosted.org/packages/6a/f7/769a0e93b5bedf8e3e791d61ca4c108bf080025e7e02017d53f5fec99832/makeprediction-4.1.1.tar.gz",
"platform": null,
"description": " \n\n<!-- \n\n-->\n\n<img src=\"assets/logo.png\" alt=\"makeprediction logo\" width=\"500px\"/>\n\n<!-- <img src=\"assets/logo_1.png\" alt=\"makeprediction logo\" width=\"300px\"/>\n\n-->\n\n \n \n\nMakePrediction is a package for building an automatic Gaussian process regression (GPR) models for time series prediction in Python. It was originally created by [Hanany Tolba].\n\n* MakePrediction is an open source project. If you have relevant skills and are interested in contributing then please do contact us (hananytolba@yahoo.com).*\n\n \n\nGaussian process regression (GPR):\n\n==========================================\n\nThe advantages of this Gaussian processes package:\n\n* Very fast training.\n\n* Very fast prediction.\n\n* The prediction can be interpolated as desired.\n\n* The training of the model is automatic: \n - *no kernel function needs to be specified*\n - *an optimal choice of kernel is automatically elaborated.*\n\n* Possibility to choose a kernel function manually.\n\n* The prediction is probabilistic (Gaussian) so that confidence intervals can be calculated and used to decide whether to make a strategic decision.\n\n* The package provides an API for deployment.\n\n \n\n## Where do you find time series?\n\n* Energy\n\n* Finance\n\n* Medical, Biotech, and Healthcare\n\n* IoT Monitoring\n\n* Supply Chain\n\n* Agriculture\n\n* Retail\n\n \n \n\n## What does makeprediction do?\n\n* Modelling and analysis time series.\n\n* Automatic time-series prediction (forecasting).\n\n* Real-Time time series prediction.\n\n* Deploy on production the fitted (or saved) makeprediction model.\n\n \n\n### Applications:\n\n* Energy consumption prediction.\n\n* Energy demand prediction.\n\n* Stock price prediction.\n\n* Stock market prediction.\n\n* ...\n## Install MakePrediction package \n\n### Latest release from PyPI\n\n\n* pip install makeprediction\n\n \n\n### Latest source from GitHub\n\n \n\n*Be aware that the `master` branch may change regularly, and new commits may break your code.*\n\n \n\n[MakePrediction GitHub repository](https://github.com/HananyTolba/MakePrediction.git), run:\n\n \n\n* pip install .\n*or*\n* python setup.py install \n\n\n\n## Get Started with MakePrediction\n\n### Example\n\n\n \n\nHere is a simple example:\n\n#### Data \n \n\n```python\nimport pandas as pd\nimport numpy as np\nimport plotly.graph_objects as go\n\nfrom makeprediction.gpts import GaussianProcessTimeSerie\nfrom makeprediction.kernels import RBF, White, Linear, Periodic, Matern\nfrom makeprediction.visualization import Visualizer\n\n \n\n#### generate a random noisy time series\n###############################\n\ndate = pd.date_range(start = '2022/06',periods = 1000, freq = '20T')\n\n# As sum of some Gaussian kernels\n\n# As sum of some Gaussian kernels\n\nkernel = RBF() + Periodic() + White(variance = .01)\n\n# add mean and variance\ndata = 100 + 10*kernel.simulate(date, seed = np.random.seed(115))\n## generate data without fixe np.random.seed\n# data = 100 + 10*kernel.simulate(date)\n\n# create a dataframe with data\ndf = pd.DataFrame(data = data, index = date, columns=['value'])\nprint(df.head())\n\n# split time serie into train and test\nTRAIN_SIZE = int(.8*len(df))\ndf_train, df_test = df[:TRAIN_SIZE], df[TRAIN_SIZE:]\n# Create an instance of the class GaussianProcessTimeSerie with train data:\n#########################################\nmodel = GaussianProcessTimeSerie(df_train.index, df_train.value)\n# Show train data with test data\nVisualizer.iplot(model, df_test.index, df_test.value)\n\n```\n\n<img src=\"assets_images/fig0.png\" width=\"900px\"/>\n\n\n\n#### Train \n\n```python\n\n# fit the model\n\nmodel.fit()\n\n```\n\n \n#### Test (long term prediction)\n\nWe will first show a simple prediction without updating with new observations (in other words without ever using df_test.value). We can say that the prediction horizon is infinite.\n\n```python\n\n\n\n#predict with model and plot result\n\nmodel.predict(df_test.index)\n\nVisualizer.iplot(model, df_test.index, df_test.value)\n\n \n\n```\n\n\n\n <img src=\"assets_images/fig1.png\" width=\"900px\"/>\n\n```python\n# Model components\n#how the components of the model (decomposition)\nfig3 = Visualizer.iplot_components(model,return_fig=True)\nfig3.write_image(\"fig3.png\",width = 900,height =700, scale = 5)\n```\n <img src=\"assets_images/fig3.png\" width=\"900px\"/>\n\n```python\n#plot only test result\n\nfig2 = Visualizer.iplot(model, df_test.index, df_test.value, return_fig=True, test_only=True)\nfig2.write_image(\"fig2.png\",width = 1400,height =700)\nfig2.show()\n\n ```\n<img src=\"assets_images/fig2.png\" width=\"900px\"/>\n\n\n#### Online prediction with horizon = 1\n\nContrary to the previous case, the horizon is 1. In other words, we predict a first value at a given time, then we will update the model by communicating this observation via the update method and so on (see codes). \n\n```python\n\n \n\n# Online prediction with updating\n\nypred = np.empty(shape = (0,))\n\nypred_std = np.empty(shape = (0,))\n\nfor x,y in df_test.itertuples():\n\n\t# predict for x value\n\n\typ,yp_std = model.predict(x,return_value = True)\n\n\typred = np.append(ypred,yp)\n\n\typred_std = np.append(ypred_std,yp_std)\n\n\t# update the model for (x,y)\n\n\tdata = {'x_update': x, 'y_update': y}\n\n\tmodel.update(**data)\n\n\n\n# Show new prediction\n\n# Show new update prediction\n\nfig2.add_trace(\n go.Scatter(\n x=df_test.index,\n y=ypred,\n mode=\"lines\",\n name='Prediction(horizon = 1)',\n showlegend=True)\n)\n\nfig2.write_image(\"fig2_update_one_head.png\",width = 900,height =700, scale = 5)\n\n \n\n\n```\n\n <img src=\"assets_images/fig2_update_one_head.png\" width=\"900px\"/>\n\n\n \n\n#### After updating, how to check?\n \nNote that each time the model is updated with a new data \tdata = {'x_update': x (time), 'y_update': y (value)}\nor even data vector. The model learns this data and adjusts itself to become more and more efficient.\n\n```python\n\n#prediction of data already seen(updated)\nyupdate, _ = model.predict(df_test.index,return_value = True)\n\nfig2.add_trace(\n go.Scatter(\n x=df_test.index,\n y=ypred,\n mode=\"lines\",\n name='Prediction(horizon = 1)',\n showlegend=True)\n)\n\n\nfig2.add_trace(\n go.Scatter(\n x=df_test.index,\n y=yupdate,\n mode=\"lines\",\n name='Update model',\n showlegend=True)\n)\nfig2.show()\nfig2.write_image(\"fig2_update.png\",width = 900,height =700, scale = 5)\n\n\n\n \n\n```\n <img src=\"assets_images/fig2_update.png\" width=\"900px\"/>\n\n \n",
"bugtrack_url": null,
"license": "GPLv3",
"summary": "Automatic and fast Gaussian process for time serie prediction.",
"version": "4.1.1",
"project_urls": {
"Download": "https://github.com/HananyTolba/MakePrediction.git",
"Homepage": "https://github.com/HananyTolba/MakePrediction.git"
},
"split_keywords": [
"gaussian process regression",
"time series prediction",
"machine learning"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "a8c5b89c5965040fedebeb9904a711102921c7f6e2b203fb3cfaf4ccbbe2f121",
"md5": "3270322bc28836c697cb15552aa45340",
"sha256": "1a506f37573c875901654f71132a136bba3162dbcb9ee269aa7d7d231cd548c3"
},
"downloads": -1,
"filename": "makeprediction-4.1.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "3270322bc28836c697cb15552aa45340",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.6",
"size": 58630,
"upload_time": "2023-05-27T11:43:13",
"upload_time_iso_8601": "2023-05-27T11:43:13.618775Z",
"url": "https://files.pythonhosted.org/packages/a8/c5/b89c5965040fedebeb9904a711102921c7f6e2b203fb3cfaf4ccbbe2f121/makeprediction-4.1.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "6af7769a0e93b5bedf8e3e791d61ca4c108bf080025e7e02017d53f5fec99832",
"md5": "94a73b8b807b6ae05dcbb515e37b9f43",
"sha256": "420fa859d5197f0292966f3a4f70b8299269b8a38724ce2b8bfbaf0c79b83a50"
},
"downloads": -1,
"filename": "makeprediction-4.1.1.tar.gz",
"has_sig": false,
"md5_digest": "94a73b8b807b6ae05dcbb515e37b9f43",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.6",
"size": 50500,
"upload_time": "2023-05-27T11:43:16",
"upload_time_iso_8601": "2023-05-27T11:43:16.336505Z",
"url": "https://files.pythonhosted.org/packages/6a/f7/769a0e93b5bedf8e3e791d61ca4c108bf080025e7e02017d53f5fec99832/makeprediction-4.1.1.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2023-05-27 11:43:16",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "HananyTolba",
"github_project": "MakePrediction",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"requirements": [],
"lcname": "makeprediction"
}
JSON
