# Perlib
This repository hosts the development of the Perlib library.
## About Perlib
Perlib is a framework written in Python where you can use deep and machine learning algorithms.
## Perlib is
Feature to use many deep or machine learning models easily
Feature to easily generate estimates in a single line with default parameters
Understanding data with simple analyzes with a single line
Feature to automatically preprocess data in a single line
Feature to easily create artificial neural networks
Feature to manually pre-process data, extract analysis or create models with detailed parameters, produce tests and predictions
## Usage
The core data structures are layers and models. For quick results with default parameters
To set up more detailed operations and structures, you should use the Perflib functional API, which allows you to create arbitrary layers or write models completely from scratch via subclassing.
## Install
```python
pip install perlib
```
```python
from perlib.forecaster import *
```
This is how you can use sample datasets.
```python
from perlib import datasets # or from perlib.datasets import *
import pandas as pd
dataset = datasets.load_airpassengers()
data = pd.DataFrame(dataset)
data.index = pd.date_range(start="2022-01-01",periods=len(data),freq="d")
```
To read your own dataset;
```python
import perlib
pr = perlib.dataPrepration()
data = pr.read_data("./datasets/winequality-white.csv",delimiter=";")
```
```python
model_selector = perlib.ModelSelection(data_l)
model_selector.select_model()
```
```python
Dimension Reduction Analysis: Size reduction is not recommended.
Selected Models:
1. ('BILSTM', 'Data symmetric, BILSTM is available.')
2. ('CONVLSTM', 'There are spatial and temporal patterns, CONVLSTM can be used.')
3. ('SARIMA', 'No time series data found, SARIMA is not recommended.')
4. ('PROPHET', 'No time series data found, PROPHET is not recommended.')
5. ('LSTM', 'Insufficient dataset size, LSTM is not recommended.')
6. ('LSTNET', 'Insufficient dataset size, LSTNet is not recommended.')
7. ('TCN', 'No temporal dependencies, TCN is not recommended.')
8. ('XGBoost', 'No irregular and heterogeneous data, XGBoost is not recommended.')
```
The easiest way to get quick results is with the 'get_result' function.
You can choice modelname ;
"RNN", "LSTM", "BILSTM", "CONVLSTM", "TCN", "LSTNET", "ARIMA" ,"SARIMA" or all machine learning algorithms
```python
forecast,evaluate = get_result(dataFrame=data,
y="Values",
modelName="Lstnet",
dateColumn=False,
process=False,
forecastNumber=24,
metric=["mape","mae","mse"],
epoch=2,
forecastingStartDate=2022-03-06
)
```
```python
Parameters created
The model training process has been started.
Epoch 1/2
500/500 [==============================] - 14s 23ms/step - loss: 0.2693 - val_loss: 0.0397
Epoch 2/2
500/500 [==============================] - 12s 24ms/step - loss: 0.0500 - val_loss: 0.0092
Model training process completed
The model is being saved
1/1 [==============================] - 0s 240ms/step
1/1 [==============================] - 0s 16ms/step
1/1 [==============================] - 0s 10ms/step
1/1 [==============================] - 0s 16ms/step
Values Predicts
Date
2022-03-07 71 79.437263
2022-03-14 84 84.282906
2022-03-21 90 88.096298
2022-03-28 87 82.875603
MAPE: 3.576822717339706
```
```python
forecast
Predicts Actual
Date
2022-03-07 71 79.437263
2022-03-08 84 84.282906
2022-03-09 90 88.096298
2022-03-10 87 82.875603
```
```python
evaluate
{'mean_absolute_percentage_error': 3.576822717339706,
'mean_absolute_error': 14.02137889193878,
'mean_squared_error': 3485.26570064559}
```
he Time Series module helps to create many basic models
without using much code and helps to understand which models
work better without any parameter adjustments.
```python
from perlib.piplines.dpipline import Timeseries
pipline = Timeseries(dataFrame=data,
y="Values",
dateColumn=False,
process=False,
epoch=1,
forecastingStartDate="2022-03-06",
forecastNumber= 24,
models="all",
metrics=["mape","mae","mse"]
)
predictions = pipline.fit()
mean_absolute_percentage_error | mean_absolute_error | mean_squared_error
LSTNET 14.05 | 67.70 | 5990.35
LSTM 7.03 | 38.28 | 2250.69
BILSTM 13.21 | 68.22 | 6661.60
CONVLSTM 9.62 | 48.06 | 2773.69
TCN 12.03 | 65.44 | 6423.10
RNN 11.53 | 59.33 | 4793.62
ARIMA 50.18 | 261.14| 74654.48
SARIMA 10.48 | 51.25 | 3238.20
```
With the 'summarize' function you can see quick and simple analysis results.
```python
summarize(dataFrame=data)
```
With the 'auto' function under 'preprocess', you can prepare the data using general preprocessing.
```python
preprocess.auto(dataFrame=data)
12-2022 15:04:36.22 - DEBUG - Conversion to DATETIME succeeded for feature "Date"
27-12-2022 15:04:36.23 - INFO - Completed conversion of DATETIME features in 0.0097 seconds
27-12-2022 15:04:36.23 - INFO - Started encoding categorical features... Method: "AUTO"
27-12-2022 15:04:36.23 - DEBUG - Skipped encoding for DATETIME feature "Date"
27-12-2022 15:04:36.23 - INFO - Completed encoding of categorical features in 0.001252 seconds
27-12-2022 15:04:36.23 - INFO - Started feature type conversion...
27-12-2022 15:04:36.23 - DEBUG - Conversion to type INT succeeded for feature "Salecount"
27-12-2022 15:04:36.24 - DEBUG - Conversion to type INT succeeded for feature "Day"
27-12-2022 15:04:36.24 - DEBUG - Conversion to type INT succeeded for feature "Month"
27-12-2022 15:04:36.24 - DEBUG - Conversion to type INT succeeded for feature "Year"
27-12-2022 15:04:36.24 - INFO - Completed feature type conversion for 4 feature(s) in 0.00796 seconds
27-12-2022 15:04:36.24 - INFO - Started validation of input parameters...
27-12-2022 15:04:36.24 - INFO - Completed validation of input parameters
27-12-2022 15:04:36.24 - INFO - AutoProcess process completed in 0.034259 seconds
```
If you want to build it yourself;
```python
from perlib.core.models.dmodels import models
from perlib.core.train import dTrain
from perlib.core.tester import dTester
```
You can use many features by calling the 'dataPrepration' function for data preparation operations.
```python
data = dataPrepration.read_data(path="./dataset/Veriler/ayakkabı_haftalık.xlsx")
```
```python
data = dataPrepration.trainingFordate_range(dataFrame=data,dt1="2013-01-01",dt2="2022-01-01")
```
You can use the 'preprocess' function for data preprocessing.
```python
data = preprocess.missing_num(dataFrame=data)
```
```python
data = preprocess.find_outliers(dataFrame=data)
```
```python
data = preprocess.encode_cat(dataFrame=data)
```
```python
data = preprocess.dublicates(dataFrame=data,mode="auto")
```
You should create an architecture like below.
```python
layers = {
"unit":[150,100],
"activation":["tanh","tanh"],
"dropout" :[0.2,0.2]
}
```
You can set each parameter below it by calling the 'req_info' object.
```python
from perlib.forecaster import req_info,dmodels
from perlib.core.train import dTrain
from perlib.core.tester import dTester
#layers = {
# "CNNFilters":100,
# "CNNKernel":6,
# "GRUUnits":50,
# "skip" : 25,
# "highway" : 1
# }
req_info.layers = None
req_info.modelname = "lstm"
req_info.epoch = 30
#req_info.learning_rate = 0.001
req_info.loss = "mse"
req_info.lookback = 30
req_info.optimizer = "adam"
req_info.targetCol = "Values"
req_info.forecastingStartDate = "2022-01-06 15:00:00"
req_info.period = "daily"
req_info.forecastNumber = 30
req_info.scaler = "standard"
s = dmodels(req_info)
```
It will be prepared after importing it into models.
```python
s = models(req_info)
```
After sending the dataframe and the prepared architecture to the dTrain, you can start the training process by calling the .fit() function.
```python
train = dTrain(dataFrame=data,object=s)
train.fit()
```
After the training is completed, you can see the results by giving the dataFrame,object,path,metric parameters to 'dTester'.
```python
t = dTester(dataFrame=data,object=s,path="Data-Lstm-2022-12-14-19-56-28.h5",metric=["mape","mae"])
```
```python
t.forecast()
```
```
1/1 [==============================] - 0s 21ms/step
1/1 [==============================] - 0s 20ms/step
1/1 [==============================] - 0s 19ms/step
1/1 [==============================] - 0s 20ms/step
1/1 [==============================] - 0s 20ms/step
1/1 [==============================] - 0s 21ms/step
1/1 [==============================] - 0s 20ms/step
1/1 [==============================] - 0s 20ms/step
1/1 [==============================] - 0s 21ms/step
1/1 [==============================] - 0s 21ms/step
1/1 [==============================] - 0s 21ms/step
1/1 [==============================] - 0s 22ms/step
1/1 [==============================] - 0s 21ms/step
```
```python
t.evaluate()
MAPE: 3.35
```
```python
from perlib.core.models.smodels import models as armodels
from perlib.core.train import sTrain
from perlib.core.tester import sTester
```
```python
aR_info.modelname = "sarima"
aR_info.forcastingStartDate = "2022-6-10"
```
```python
ar = armodels(aR_info)
#train = sTrain(dataFrame=data,object=ar)
res = train.fit()
```
```python
r = sTester(dataFrame=data,object=ar,path="Data-sarima-2022-12-30-23-49-03.pkl")
r.forecast()
r.evaluate()
```
```python
from perlib.core.models.mmodels import models
from perlib.core.train import mTrain
```
```python
m_info.testsize = .01
m_info.y = "quality"
m_info.modelname= "SVR"
m_info.auto = False
```
```python
m = models(m_info)
train = mTrain(dataFrame=data,object=m)
preds, evaluate = train.predict()
```
```python
# If you want to make any other data predictions you can use the train.tester
# func after train.predict. You can make predictions with
predicts = train.tester(path="Data-SVR-2023-01-08-09-50-37.pkl", testData=data.iloc[:,1:][-20:])
```
Raw data
{
"_id": null,
"home_page": "https://github.com/Ruzzg/perlib",
"name": "perlib",
"maintainer": "",
"docs_url": null,
"requires_python": "",
"maintainer_email": "",
"keywords": "perlib,tensorflow,machine learning,deep learning,statistical learning,automl,autodl,modelselection",
"author": "R\u00fczgar Ersin Kanar",
"author_email": "ruzgarknr@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/9d/dc/549bf5c060c7d22907181a8c092aafa541c09ebe9a94c713b6738c73e46f/perlib-3.0.4.tar.gz",
"platform": null,
"description": "\n\n\n# Perlib\n\nThis repository hosts the development of the Perlib library.\n\n## About Perlib\n\nPerlib is a framework written in Python where you can use deep and machine learning algorithms.\n\n## Perlib is\nFeature to use many deep or machine learning models easily\nFeature to easily generate estimates in a single line with default parameters\nUnderstanding data with simple analyzes with a single line\nFeature to automatically preprocess data in a single line\nFeature to easily create artificial neural networks\nFeature to manually pre-process data, extract analysis or create models with detailed parameters, produce tests and predictions\n\n## Usage\nThe core data structures are layers and models. For quick results with default parameters\nTo set up more detailed operations and structures, you should use the Perflib functional API, which allows you to create arbitrary layers or write models completely from scratch via subclassing.\n\n\n## Install\n```python\npip install perlib\n```\n\n```python\nfrom perlib.forecaster import *\n```\n\nThis is how you can use sample datasets.\n\n```python\nfrom perlib import datasets # or from perlib.datasets import *\nimport pandas as pd\ndataset = datasets.load_airpassengers()\ndata = pd.DataFrame(dataset)\ndata.index = pd.date_range(start=\"2022-01-01\",periods=len(data),freq=\"d\")\n```\n\nTo read your own dataset;\n```python \nimport perlib\npr = perlib.dataPrepration()\ndata = pr.read_data(\"./datasets/winequality-white.csv\",delimiter=\";\")\n```\n\n```python \nmodel_selector = perlib.ModelSelection(data_l)\nmodel_selector.select_model()\n```\n\n\n```python\n\nDimension Reduction Analysis: Size reduction is not recommended.\n\nSelected Models:\n1. ('BILSTM', 'Data symmetric, BILSTM is available.')\n2. ('CONVLSTM', 'There are spatial and temporal patterns, CONVLSTM can be used.')\n3. ('SARIMA', 'No time series data found, SARIMA is not recommended.')\n4. ('PROPHET', 'No time series data found, PROPHET is not recommended.')\n5. ('LSTM', 'Insufficient dataset size, LSTM is not recommended.')\n6. ('LSTNET', 'Insufficient dataset size, LSTNet is not recommended.')\n7. ('TCN', 'No temporal dependencies, TCN is not recommended.')\n8. ('XGBoost', 'No irregular and heterogeneous data, XGBoost is not recommended.')\n\n```\n\n\nThe easiest way to get quick results is with the 'get_result' function.\nYou can choice modelname ;\n\"RNN\", \"LSTM\", \"BILSTM\", \"CONVLSTM\", \"TCN\", \"LSTNET\", \"ARIMA\" ,\"SARIMA\" or all machine learning algorithms\n\n\n```python \nforecast,evaluate = get_result(dataFrame=data,\n y=\"Values\",\n modelName=\"Lstnet\",\n dateColumn=False,\n process=False,\n forecastNumber=24,\n metric=[\"mape\",\"mae\",\"mse\"],\n epoch=2,\n forecastingStartDate=2022-03-06\n )\n```\n```python \n\nParameters created\nThe model training process has been started.\nEpoch 1/2\n500/500 [==============================] - 14s 23ms/step - loss: 0.2693 - val_loss: 0.0397\nEpoch 2/2\n500/500 [==============================] - 12s 24ms/step - loss: 0.0500 - val_loss: 0.0092\nModel training process completed\n\nThe model is being saved\n1/1 [==============================] - 0s 240ms/step\n1/1 [==============================] - 0s 16ms/step\n1/1 [==============================] - 0s 10ms/step\n1/1 [==============================] - 0s 16ms/step\n Values Predicts\nDate \n2022-03-07 71 79.437263\n2022-03-14 84 84.282906\n2022-03-21 90 88.096298\n2022-03-28 87 82.875603\nMAPE: 3.576822717339706\n\n```\n\n```python \nforecast\n\n Predicts Actual\nDate \n2022-03-07 71 79.437263\n2022-03-08 84 84.282906\n2022-03-09 90 88.096298\n2022-03-10 87 82.875603\n```\n\n```python \nevaluate\n\n{'mean_absolute_percentage_error': 3.576822717339706,\n 'mean_absolute_error': 14.02137889193878,\n 'mean_squared_error': 3485.26570064559}\n```\n\n\nhe Time Series module helps to create many basic models\nwithout using much code and helps to understand which models \nwork better without any parameter adjustments.\n```python \nfrom perlib.piplines.dpipline import Timeseries\npipline = Timeseries(dataFrame=data,\n y=\"Values\",\n dateColumn=False,\n process=False,\n epoch=1,\n forecastingStartDate=\"2022-03-06\",\n forecastNumber= 24,\n models=\"all\",\n metrics=[\"mape\",\"mae\",\"mse\"]\n )\npredictions = pipline.fit()\n\n mean_absolute_percentage_error | mean_absolute_error | mean_squared_error\nLSTNET 14.05 | 67.70 | 5990.35\nLSTM 7.03 | 38.28 | 2250.69\nBILSTM 13.21 | 68.22 | 6661.60\nCONVLSTM 9.62 | 48.06 | 2773.69\nTCN 12.03 | 65.44 | 6423.10\nRNN 11.53 | 59.33 | 4793.62\nARIMA 50.18 | 261.14| 74654.48\nSARIMA 10.48 | 51.25 | 3238.20\n```\n\n\nWith the 'summarize' function you can see quick and simple analysis results.\n```python \nsummarize(dataFrame=data)\n```\n\n\nWith the 'auto' function under 'preprocess', you can prepare the data using general preprocessing.\n```python \npreprocess.auto(dataFrame=data)\n\n12-2022 15:04:36.22 - DEBUG - Conversion to DATETIME succeeded for feature \"Date\"\n27-12-2022 15:04:36.23 - INFO - Completed conversion of DATETIME features in 0.0097 seconds\n27-12-2022 15:04:36.23 - INFO - Started encoding categorical features... Method: \"AUTO\"\n27-12-2022 15:04:36.23 - DEBUG - Skipped encoding for DATETIME feature \"Date\"\n27-12-2022 15:04:36.23 - INFO - Completed encoding of categorical features in 0.001252 seconds\n27-12-2022 15:04:36.23 - INFO - Started feature type conversion...\n27-12-2022 15:04:36.23 - DEBUG - Conversion to type INT succeeded for feature \"Salecount\"\n27-12-2022 15:04:36.24 - DEBUG - Conversion to type INT succeeded for feature \"Day\"\n27-12-2022 15:04:36.24 - DEBUG - Conversion to type INT succeeded for feature \"Month\"\n27-12-2022 15:04:36.24 - DEBUG - Conversion to type INT succeeded for feature \"Year\"\n27-12-2022 15:04:36.24 - INFO - Completed feature type conversion for 4 feature(s) in 0.00796 seconds\n27-12-2022 15:04:36.24 - INFO - Started validation of input parameters...\n27-12-2022 15:04:36.24 - INFO - Completed validation of input parameters\n27-12-2022 15:04:36.24 - INFO - AutoProcess process completed in 0.034259 seconds\n```\n\n\n\nIf you want to build it yourself;\n```python \nfrom perlib.core.models.dmodels import models\nfrom perlib.core.train import dTrain\nfrom perlib.core.tester import dTester\n```\n You can use many features by calling the 'dataPrepration' function for data preparation operations.\n```python \ndata = dataPrepration.read_data(path=\"./dataset/Veriler/ayakkab\u0131_haftal\u0131k.xlsx\")\n```\n```python \ndata = dataPrepration.trainingFordate_range(dataFrame=data,dt1=\"2013-01-01\",dt2=\"2022-01-01\")\n```\n\nYou can use the 'preprocess' function for data preprocessing.\n```python \ndata = preprocess.missing_num(dataFrame=data)\n```\n```python \ndata = preprocess.find_outliers(dataFrame=data)\n```\n```python \ndata = preprocess.encode_cat(dataFrame=data)\n```\n```python \ndata = preprocess.dublicates(dataFrame=data,mode=\"auto\")\n```\n\nYou should create an architecture like below.\n```python \nlayers = {\n \"unit\":[150,100], \n \"activation\":[\"tanh\",\"tanh\"],\n \"dropout\" :[0.2,0.2]\n }\n```\n\nYou can set each parameter below it by calling the 'req_info' object.\n```python \nfrom perlib.forecaster import req_info,dmodels\nfrom perlib.core.train import dTrain\nfrom perlib.core.tester import dTester\n\n#layers = {\n# \"CNNFilters\":100,\n# \"CNNKernel\":6,\n# \"GRUUnits\":50,\n# \"skip\" : 25,\n# \"highway\" : 1\n# }\n\nreq_info.layers = None\nreq_info.modelname = \"lstm\"\nreq_info.epoch = 30\n#req_info.learning_rate = 0.001\nreq_info.loss = \"mse\"\nreq_info.lookback = 30\nreq_info.optimizer = \"adam\"\nreq_info.targetCol = \"Values\"\nreq_info.forecastingStartDate = \"2022-01-06 15:00:00\"\nreq_info.period = \"daily\"\nreq_info.forecastNumber = 30\nreq_info.scaler = \"standard\"\ns = dmodels(req_info)\n\n```\n\nIt will be prepared after importing it into models.\n```python \ns = models(req_info)\n```\n\nAfter sending the dataframe and the prepared architecture to the dTrain, you can start the training process by calling the .fit() function.\n```python \ntrain = dTrain(dataFrame=data,object=s)\ntrain.fit()\n```\nAfter the training is completed, you can see the results by giving the dataFrame,object,path,metric parameters to 'dTester'.\n```python \nt = dTester(dataFrame=data,object=s,path=\"Data-Lstm-2022-12-14-19-56-28.h5\",metric=[\"mape\",\"mae\"])\n```\n```python \nt.forecast()\n```\n```\n1/1 [==============================] - 0s 21ms/step\n1/1 [==============================] - 0s 20ms/step\n1/1 [==============================] - 0s 19ms/step\n1/1 [==============================] - 0s 20ms/step\n1/1 [==============================] - 0s 20ms/step\n1/1 [==============================] - 0s 21ms/step\n1/1 [==============================] - 0s 20ms/step\n1/1 [==============================] - 0s 20ms/step\n1/1 [==============================] - 0s 21ms/step\n1/1 [==============================] - 0s 21ms/step\n1/1 [==============================] - 0s 21ms/step\n1/1 [==============================] - 0s 22ms/step\n1/1 [==============================] - 0s 21ms/step\n\n```\n\n```python \nt.evaluate()\n\nMAPE: 3.35\n```\n\n```python \nfrom perlib.core.models.smodels import models as armodels\nfrom perlib.core.train import sTrain\nfrom perlib.core.tester import sTester\n```\n```python \naR_info.modelname = \"sarima\"\naR_info.forcastingStartDate = \"2022-6-10\"\n```\n```python \nar = armodels(aR_info)\n#train = sTrain(dataFrame=data,object=ar)\nres = train.fit()\n```\n```python \nr = sTester(dataFrame=data,object=ar,path=\"Data-sarima-2022-12-30-23-49-03.pkl\")\nr.forecast()\nr.evaluate()\n```\n\n```python \nfrom perlib.core.models.mmodels import models\nfrom perlib.core.train import mTrain\n```\n\n```python \nm_info.testsize = .01\nm_info.y = \"quality\"\nm_info.modelname= \"SVR\"\nm_info.auto = False\n```\n\n```python \nm = models(m_info)\ntrain = mTrain(dataFrame=data,object=m)\npreds, evaluate = train.predict()\n```\n\n```python \n# If you want to make any other data predictions you can use the train.tester\n# func after train.predict. You can make predictions with\npredicts = train.tester(path=\"Data-SVR-2023-01-08-09-50-37.pkl\", testData=data.iloc[:,1:][-20:])\n```\n\n\n",
"bugtrack_url": null,
"license": "Apache Software License",
"summary": "Deep learning, Machine learning and Statistical learning for humans.",
"version": "3.0.4",
"project_urls": {
"Homepage": "https://github.com/Ruzzg/perlib"
},
"split_keywords": [
"perlib",
"tensorflow",
"machine learning",
"deep learning",
"statistical learning",
"automl",
"autodl",
"modelselection"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "9ddc549bf5c060c7d22907181a8c092aafa541c09ebe9a94c713b6738c73e46f",
"md5": "bdd336350f9681033f7ab0dfc1cec5cf",
"sha256": "565de831430610040aa4b8689081cb67a042be34e184cb9678cb58639dd524c0"
},
"downloads": -1,
"filename": "perlib-3.0.4.tar.gz",
"has_sig": false,
"md5_digest": "bdd336350f9681033f7ab0dfc1cec5cf",
"packagetype": "sdist",
"python_version": "source",
"requires_python": null,
"size": 136132,
"upload_time": "2024-03-19T17:44:53",
"upload_time_iso_8601": "2024-03-19T17:44:53.542574Z",
"url": "https://files.pythonhosted.org/packages/9d/dc/549bf5c060c7d22907181a8c092aafa541c09ebe9a94c713b6738c73e46f/perlib-3.0.4.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-03-19 17:44:53",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "Ruzzg",
"github_project": "perlib",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "perlib"
}
JSON
