Generate.py

import sklearn
import pandas as pd
from tsai.basics import *
import config
from tsai.inference import load_learner

import pandas as pd       



def get_inputs_from_user():
    
    return 0

def preprocess_data(DataFrame:pd.DataFrame):
    preproc=load_object()        
    return DataFrame

def preprocess_data_transform_generate_splits_Train(DataFrame:pd.DataFrame):
    DataFrame=DataFrame.drop(config.DROP_COLOUMNS,axis=1)
    preproc_pipe=load_object(config.PREPROCESSOR_PATH)
    exp_pipe=load_object(config.SCALING_DATA)
    DataFrame=preproc_pipe.fit_transform(DataFrame)
    
    print("dataframe processed and ready for splitting")
    
    splits=get_forecasting_splits(DataFrame,fcst_history=config.FCST_HISTORY,fcst_horizon=config.FCST_HORIZON,datetime_col=config.DATETIME_COL,
                                  valid_size=config.VALID_SIZE,test_size=config.TEST_SIZE)
    
    X,y=prepare_forecasting_data(DataFrame,fcst_history=config.FCST_HISTORY,fcst_horizon=config.FCST_HORIZON,x_vars=config.COLOUMNS,y_vars=config.COLOUMNS)
    
    learn=TSForecaster(X,y,splits=splits,
        batch_size=16,path='models',
        arch='InceptionTimePlus',#"PatchTST" when PatchTST is to be used
        pipelines=[preproc_pipe,exp_pipe],
        #arch_config=config.ARCH_CONFIG, #uncomment only if PatchTST is used
        metrics=[mse,mape],
        cbs=ShowGraph()
    )
    
    lr_max=learn.lr_find().valley
    
    learn.fit_one_cycle(n_epoch=config.N_EPOCH,lr_max=lr_max)
    learn.export("model_in.pt")
    return 0

#when using PatchTst model use the below function 
def inference_Aircomp(fcst_date:string,DataFrame:pd.DataFrame):
    pre=load_object(config.AIR_PREPROCESSOR_PATH)
    DataFrame=pre.fit_transform(DataFrame)
    
    dates=pd.date_range(start=None,end=fcst_date,periods=config.FCST_HISTORY,freq=config.FREQUENCY)
    new_df=DataFrame[DataFrame[config.AIR_DATETIME_COL].isin(dates)].reset_index(drop=True)
    

    predict=load_learner(config.MODEL_PATH_ITP_AIR)
    new_df=predict.transform(new_df)
    
    new_x,__=prepare_forecasting_data(new_df,fcst_history=config.FCST_HISTORY,fcst_horizon=0,x_vars=config.AIR_COLOUMNS,y_vars=config.AIR_COLOUMNS)
    
    new_scaled_preds, *_ = predict.get_X_preds(new_x)

    new_scaled_preds=to_np(new_scaled_preds).swapaxes(1,2).reshape(-1,len(config.AIR_COLOUMNS))
    
    dates=pd.date_range(start=fcst_date, periods=config.FCST_HORIZON+1,freq='1H')[1:]
    preds_df=pd.DataFrame(dates,columns=[config.AIR_DATETIME_COL])
    preds_df.loc[:, config.AIR_COLOUMNS]=new_scaled_preds
    preds_df=predict.inverse_transform(preds_df)
    
    return preds_df

def inference_Energy(fcst_date:string,DataFrame:pd.DataFrame):
    pre=load_object(config.ENER_PREPROCESSOR_PATH)
    DataFrame[config.ENERGY_DATETIME]=pd.to_datetime(DataFrame[config.ENERGY_DATETIME],format='mixed')   
    DataFrame=pre.fit_transform(DataFrame)
    
    dates=pd.date_range(start=None,end=fcst_date,periods=config.FCST_HISTORY,freq=config.FREQUENCY)
    new_df=DataFrame[DataFrame[config.ENERGY_DATETIME].isin(dates)].reset_index(drop=True)
    

    predict=load_learner(config.MODEL_PATH_ITP_ENER)
    new_df=predict.transform(new_df)
    
    new_x,__=prepare_forecasting_data(new_df,fcst_history=config.FCST_HISTORY,fcst_horizon=0,x_vars=config.ENERGY_COLOUMNS,y_vars=config.ENERGY_COLOUMNS)
    
    new_scaled_preds, *_ = predict.get_X_preds(new_x)

    new_scaled_preds=to_np(new_scaled_preds).swapaxes(1,2).reshape(-1,len(config.ENERGY_COLOUMNS))
    
    dates=pd.date_range(start=fcst_date, periods=config.FCST_HORIZON+1,freq='1H')[1:]
    preds_df=pd.DataFrame(dates,columns=[config.ENERGY_DATETIME])
    preds_df.loc[:, config.ENERGY_COLOUMNS]=new_scaled_preds
    preds_df=predict.inverse_transform(preds_df)
    
    return preds_df

def inference_boiler(fcst_date:string,DataFrame:pd.DataFrame):
    pre=load_object(config.BOILER_PREPROCESSOR_PATH)  
    DataFrame=pre.fit_transform(DataFrame)
    
    dates=pd.date_range(start=None,end=fcst_date,periods=config.FCST_HISTORY,freq=config.FREQUENCY)
    new_df=DataFrame[DataFrame[config.BOILER_DATETIME].isin(dates)].reset_index(drop=True)

    
    predict=load_learner(config.MODEL_PATH_ITP_BOIL)
    new_df=predict.transform(new_df)
    
    new_x,__=prepare_forecasting_data(new_df,fcst_history=config.FCST_HISTORY,fcst_horizon=0,x_vars=config.BOILER_COLOUMNS,y_vars=config.BOILER_COLOUMNS)
    
    new_scaled_preds, *_ = predict.get_X_preds(new_x)

    new_scaled_preds=to_np(new_scaled_preds).swapaxes(1,2).reshape(-1,len(config.BOILER_COLOUMNS))
    
    dates=pd.date_range(start=fcst_date, periods=config.FCST_HORIZON+1,freq='1H')[1:]
    preds_df=pd.DataFrame(dates,columns=[config.BOILER_DATETIME])
    preds_df.loc[:, config.BOILER_COLOUMNS]=new_scaled_preds
    preds_df=predict.inverse_transform(preds_df)
    
    return preds_df