MATLAB用拟合出的代码绘图-predictive_control:预测控制

  • p8_869938
    了解作者
  • 164.6MB
    文件大小
  • zip
    文件格式
  • 0
    收藏次数
  • VIP专享
    资源类型
  • 0
    下载次数
  • 2022-04-30 03:29
    上传日期
MATLAB用拟合出的代码绘图这是我们论文中使用的Matlab代码:Leone,Postel,Mary,Fraisse,Vallee,Viader,de La Sayette,Peschanski,Dayan,Eustache,Gagnepain *预测遭受创伤后的未来,以压制过去 *通讯作者: 该代码由4个文件夹组成: ================== SIMULATION文件夹 模型伪造 这些模拟的目的是确定我们的计算模型(请参阅我们的论文以获取模型描述)是否能够生成我们通常在TNT任务的四个块中观察到的行为减少的入侵比例(见图2)。 与我们的真实实验一样,我们设计了一个虚拟实验设置,在4个TNT会话中分配了144个抑制提示。 我们从第一个试验的.5信念开始,在每个新的模拟试验中,我们根据考虑的感知模型并随机绘制相应的感知参数生成了一个新的信念。 引入了一个抑制参数来模拟记忆抑制并避免信念轨迹向1倾斜。 要启动模型伪造,请访问/simulation/core/model_falsification.m。 模型伪造的结果存储在/simulation/store/simulatio
predictive_control-main.zip
内容介绍
This is the Matlab code used for our paper: Predicting the future to suppress the past after trauma, by Leone, Postel, Mary, Fraisse, Vallee, Viader, de La Sayette, Peschanski, Dayan, Eustache, Gagnepain* *corresponding author: pierre.gagnepain@inserm.fr This code is composed of 4 folders: =================== SIMULATION folder ------------------- - MODEL FALSIFICATION The goal of these simulations was to establish whether our computational models (please see our paper for model description) were able to generate the behavioral reduction in intrusion proportion that we normally observe across the four blocks of the TNT task (see Fig. 2). We designed a virtual experimental setting with 144 suppression cues distributed across 4 TNT sessions, as in our real experiment. We started with a belief of .5 for the first trial, and at each new simulated trial, we generated a new belief based on the perceptual model considered and randomly drawn corresponding perceptual parameters. A suppression parameter was introduced to simulate memory suppression and to avoid the tilting of belief trajectories toward 1. 1) To launch the model falsification, please see /simulation/core/model_falsification.m. 2) The outcomes of model falsification is stored in /simulation/store/simulation_belief2intrusion_beta_vfinale_v1.mat: - We simulated 200 virtual participants (each of this virtual participant is encoded into a cell structure in simulation_belief2intrusion_beta_vfinale_v1) and repeated the virtual experiment 100 times using perceptual parameter randomly drawn from a Gaussian priors distribution tailored to match our own data (to sample plausible parameters) - Binary rating generated for each of these 200 simulations were averaged across repeated sampling and summarized as intrusion proportion across the 4 artificial TNT sessions. 3) We then compared these simulated intrusion proportion with real data. Comparaison plot used in figure 2 of our paper can be reproduced using /simulation/pipeline_fig2/figure2.m - PARAMETER & MODEL RECOVERY This section tests the ability of our models to recover their parameters (parameter recovery), and verify the reliability of the model selection criterion for identifying the true generative model within a set of competitive models, ensuring that this selection is not biased in favor of one particular model (model recovery). Parameter recovery tests the generative performance of a model, by verifying whether the fitting procedure produces meaningful parameters, namely the true parameters used to generate the data. We fitted the different models to the synthetic data (generated during model falsification), in order estimate the free parameters and compute the correlation with the parameter that generated the data. This correlation was computed for each of the 200 virtual participants, using 100 randomly sampled free parameters 1) To launch parameter and model recovery, please see /simulation/core/parameter_and_model_recovery.m. 2) The outcomes of these analyses is stored in /simulation/result_simulation (already stored). - Each of the 200 virtual participants are stored as "simulation_belief2intrusion_beta_vfinale_v1_XX" with XX being the virtual participant index - This file contains a structure name "fittedsim" with estimated perceptual parameter and Log-model evidence (LME) for simulated and fitted models. - Please see /simulation/pipeline_fig2/figure2.m to plot the outcome of this analysis - Please note that each of the 200 simulation takes time to run and should not be launch on the same matlab. We use this code with our computation grid and queue submission system to launch the 200 jobs in parallel. ======================== COMPUTATIONAL DCM folder ------------------------ This folder contains the code to run computational DCM (requires SPM12 on your matlab path !!!) 1) The main starting code is start_computational_dcm.m. 2) This function requires your own data, but we provide one exemple in "REMEMBEREX001" folder. 3) This function will: - Define the DCM voi in native space, using the MNI coordinates defined in "roicoordinate.mat" - Estimate the belief trajectories from intrusion rating using the specified perceptual model (here tapas hgf 2 levels) - Run computational-DCM, using these trajectories as parametric modulator of the DCM modulatory stick function ========================== MODEL_FIT folder -------------------------- Contains the main function to estimate "State", "Item", and "Combined" belief trajectories using intrusion rating given a perceptual and an observation model (tapas_binary_combined.m) ========================== TAPAS_DEPENDENCIES -------------------------- tapas functions used during model configuration and estimation
评论
    相关推荐