0

我想使用 gluonTS 包中的 DeepAREstimator 创建预测模型。如何使用 Ray 进行超参数调优？这是示例代码。

!pip install --upgrade mxnet-cu101==1.6.0.post0
!pip install --upgrade mxnet==1.6.0
!pip install gluonts


import pandas as pd
import matplotlib as mpl
import matplotlib.pyplot as plt
from gluonts.model.deepar import DeepAREstimator
from gluonts.mx.trainer import Trainer
import numpy as np
from gluonts.dataset.common import ListDataset
from gluonts.dataset.field_names import FieldName


#Download data
!wget https://archive.ics.uci.edu/ml/machine-learning-databases/00321/LD2011_2014.txt.zip
!unzip LD2011_2014.txt.zip

df=pd.read_csv('LD2011_2014.txt', sep=';', index_col=0, parse_dates=True, decimal=',')
df_input=df.reset_index(drop=True).T.reset_index()
ts_code=df_input["index"].astype('category').cat.codes.values

#Split to train and test
df_train=df_input.iloc[:,1:134999].values
df_test=df_input.iloc[:,134999:].values

freq="15min"
start_train = pd.Timestamp("2011-01-01 00:15:00", freq=freq)
start_test = pd.Timestamp("2014-11-07 05:30:00", freq=freq)
prediction_lentgh=672
estimator = DeepAREstimator(freq=freq, 
                            context_length=672,
                            prediction_length=prediction_lentgh,
                            use_feat_static_cat=True,
                            cardinality=[1],
                            num_layers=2,
                            num_cells=32,
                            cell_type='lstm',
                            trainer=Trainer(epochs=5))
                            
train_ds = ListDataset([
    {
        FieldName.TARGET: target,
        FieldName.START: start_train,
        FieldName.FEAT_STATIC_CAT: fsc
    }
    for (target, fsc) in zip(df_train,
                             ts_code.reshape(-1,1))
], freq=freq)

test_ds = ListDataset([
    {
        FieldName.TARGET: target,
        FieldName.START: start_test,
        FieldName.FEAT_STATIC_CAT: fsc
    }
    for (target, fsc) in zip(df_test,
                            ts_code.reshape(-1,1))
], freq=freq)

predictor = estimator.train(training_data=train_ds)
4

0 回答 0