EnergyModels 0.0.5

Prediction models in timeseries

Energy Models Package

                         THIS IS A PACKAGE OF MODELS OF PREDICT IN TIMESERIES FORECASTING                
         this package helps any developer in univariate and multivariate-multi-step time series forcasting in house-power-consumption dataset lets take a look about each type 
         Real-world time series forecasting is challenging for a whole host of reasons not limited to problem features such as having multiple input variables,the requirement 
         to predict multiple time steps,nd the need to a perform the same type of prediction for multiple physical sites.

Installation

pip install EnergyModels

Models list

• LSTM
• LSTM-CNN
• BILSTM
• GRU
• BIGRU
• TimeDistributer
• CNN
• TCN
• Transformer
• Performer

All models take 3 parameters except

• TCN :

  * must take value 
    * 1 : n_steps
    * 2 : n_features 
  * default value = 1 
    * 3 : n_outputs  
  
  **TCN Model you can build it by just give it data to build function**
  

• Transformer :

  * must take value 
    * 1 : input_shape
    * 2 : n_outputs
    * 3 : head_size
    * 4 : num_heads
    * 5 : ff_dim
    * 6 : num_transformer_blocks
    * 7 : mlp_units 
  * default value = 0 
    * 8 , 9 : dropout  , mlp_units
  

• Performer :

    * must take value 
      * 1 : maxlen
      * 2 : n_features
      * 3 : n_outputs
      * 4 : vocab_size
      * 5 : embed_dim
      * 6 : num_heads
      * 7 : ff_dim 
    * default value:
      * 8 : method => 'linear'
      * 9 : supports => 10
      * 10 : rate => 0.1 
  

Package Folders

• Data
• Energy_Models

how to use the package

first you must read the data set you want to use the models on it and then import preprocess_data from Data folder :

from Data import preprocess_data as pr

df = pd.read_csv('Data.txt',sep=';', 
                 parse_dates={'date_time' : ['Date', 'Time']}, infer_datetime_format=True, 
                 low_memory=False, na_values=['nan','?'], index_col='date_time')

pr.fill_missing(df.values)
df.to_csv('new_data.csv')

df = pd.read_csv('new_data.csv',parse_dates=['date_time'], index_col= 'date_time')

next step you can use functions on preprocess_data to split and scale the data .

X_train, X_test = pr.train_test_split(df)
X_train, X_test, scaler = pr.scale_data(X_train, X_test)

After that converting the data to supervised. now you can build model by import it from models folder :

from Energy_Models import * as m
model=m.lstm(21, 7 , 7 ).getModel()
21 ==> n_steps
7 ==> n_features
7==>n_outputs

After that you will able to predict and evaluate your models used.

y=model.predict(X)
X==>input

now you can calculate loss using metrics function for train and test both by just primt_metrix func :

exists on Evaluation_Metrix if u want to just import it :

from Energy_Models import Evaluation_Metrix as mx

mx.print_metrics(Y_train,Y_pred_train,Y_test,Y_pred_test)