如何在Synthcity库的TimeGAN模型中无结果变量完成训练？

要在没有结果特征的情况下训练TimeGAN，你可以通过以下两种方式处理TimeSeriesDataLoader的outcome参数：

方法一：显式传入outcome=None #

TimeSeriesDataLoader的outcome参数支持传入None来表示没有结果特征，只需在实例化时明确指定该参数即可，无需省略：

# instantiate time series data loader
loader = TimeSeriesDataLoader(
    temporal_data=temporal_dataframes,
    observation_times=observation_data,
    static_data=static_data,
    outcome=None  # 明确声明无结果特征
)
plugin_params = dict(
            n_iter = 20,
            batch_size = 100,
            lr = 0.001,
            generator_n_layers_hidden = 2,
            generator_n_units_hidden = 2,
            discriminator_n_layers_hidden = 2,
            discriminator_n_units_hidden = 2
        )
syn_model = Plugins().get("timegan", **plugin_params)

syn_model.fit(loader)

方法二：构造空的结果特征集合 #

如果传入None仍报错，可以构造一个空的结果数据结构（比如空DataFrame或空数组），确保样本数量与其他数据匹配：

import pandas as pd

# 创建空的结果特征DataFrame
outcome_data = pd.DataFrame()

# instantiate time series data loader
loader = TimeSeriesDataLoader(
    temporal_data=temporal_dataframes,
    observation_times=observation_data,
    static_data=static_data,
    outcome=outcome_data
)
plugin_params = dict(
            n_iter = 20,
            batch_size = 100,
            lr = 0.001,
            generator_n_layers_hidden = 2,
            generator_n_units_hidden = 2,
            discriminator_n_layers_hidden = 2,
            discriminator_n_units_hidden = 2
        )
syn_model = Plugins().get("timegan", **plugin_params)

syn_model.fit(loader)

TimeGAN的核心训练逻辑依赖时间序列和静态特征，空的结果特征不会对模型训练产生影响，仅用于满足加载器的参数格式要求。

内容的提问来源于stack exchange，提问作者Leyla Elkhamlichi