diffusion.py

import numpy as np
import torch
from torch import nn
import torch.nn.functional as F
import pdb

# import diffuser.utils as utils
from helpers import (
    cosine_beta_schedule,
    extract,
    apply_conditioning,
    Losses,
)

class GaussianDiffusion(nn.Module):
    def __init__(self, model, horizon, observation_dim, action_dim, n_timesteps=1000,
                 loss_type='l1', clip_denoised=False, predict_epsilon=True,
                 action_weight=1.0, loss_discount=1.0, loss_weights=None, returns_condition=False,
                 condition_guidance_w=0.1,):
        super().__init__()
        self.horizon = horizon
        self.observation_dim = observation_dim
        self.action_dim = action_dim
        self.transition_dim = observation_dim + action_dim
        self.model = model
        self.returns_condition = returns_condition
        self.condition_guidance_w = condition_guidance_w

        betas = cosine_beta_schedule(n_timesteps, 1e-4)
        alphas = 1. - betas
        alphas_cumprod = torch.cumprod(alphas, axis=0)
        alphas_cumprod_prev = torch.cat([torch.ones(1), alphas_cumprod[:-1]])

        self.n_timesteps = int(n_timesteps)
        self.clip_denoised = clip_denoised
        self.predict_epsilon = predict_epsilon

        
        self.register_buffer('betas', betas)
        self.register_buffer('alphas_cumprod', alphas_cumprod)
        self.register_buffer('alphas_cumprod_prev', alphas_cumprod_prev)

        # calculations for diffusion q(x_t | x_{t-1}) and others
        self.register_buffer('sqrt_alphas_cumprod', torch.sqrt(alphas_cumprod))
        self.register_buffer('sqrt_one_minus_alphas_cumprod', torch.sqrt(1. - alphas_cumprod))
        self.register_buffer('log_one_minus_alphas_cumprod', torch.log(1. - alphas_cumprod))
        self.register_buffer('sqrt_recip_alphas_cumprod', torch.sqrt(1. / alphas_cumprod))
        self.register_buffer('sqrt_recipm1_alphas_cumprod', torch.sqrt(1. / alphas_cumprod - 1))

        # calculations for posterior q(x_{t-1} | x_t, x_0)
        posterior_variance = betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod)
        self.register_buffer('posterior_variance', posterior_variance)

        ## log calculation clipped because the posterior variance
        ## is 0 at the beginning of the diffusion chain
        self.register_buffer('posterior_log_variance_clipped',
            torch.log(torch.clamp(posterior_variance, min=1e-20)))
        self.register_buffer('posterior_mean_coef1',
            betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod))
        self.register_buffer('posterior_mean_coef2',
            (1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod))
        
        ## get loss coefficients and initialize objective
        loss_weights = self.get_loss_weights(action_weight, loss_discount, loss_weights)
        self.loss_fn = Losses[loss_type](loss_weights, self.action_dim)
    
    def get_loss_weights(self, action_weight, discount, weights_dict):
        '''
            sets loss coefficients for trajectory

            action_weight   : float
                coefficient on first action loss
            discount   : float
                multiplies t^th timestep of trajectory loss by discount**t
            weights_dict    : dict
                { i: c } multiplies dimension i of observation loss by c
        '''
        self.action_weight = action_weight
        dim_weights = torch.ones(self.transition_dim, dtype=torch.float32)

        ## set loss coefficients for dimensions of observation
        if weights_dict is None: weights_dict = {}
        for ind, w in weights_dict.items():
            dim_weights[self.action_dim + ind] *= w

        ## decay loss with trajectory timestep: discount**t
        discounts = discount ** torch.arange(self.horizon, dtype=torch.float)
        discounts = discounts / discounts.mean()
        loss_weights = torch.einsum('h,t->ht', discounts, dim_weights)

        ## manually set a0 weight
        loss_weights[0, :self.action_dim] = action_weight
        return loss_weights
    
    #------------------------------------------ sampling ------------------------------------------#
    def predict_start_from_noise(self,x_t, t, noise):
        '''
            if self.predict_epsilon, model output is (scaled) noise;
            otherwise, model predicts x0 directly
        '''
        if self.predict_epsilon:
            return (
                extract(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
                extract(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
            )
        else:
            return noise
    
    def q_posterior(self, x_start, x_t, t):
        '''
            q(x_{t-1} | x_t, x_0)
        '''
        posterior_mean = (
            extract(self.posterior_mean_coef1, t, x_t.shape) * x_start +
            extract(self.posterior_mean_coef2, t, x_t.shape) * x_t
        )
        posterior_variance = extract(self.posterior_variance, t, x_t.shape)
        posterior_log_variance_clipped = extract(self.posterior_log_variance_clipped, t, x_t.shape)
        return posterior_mean, posterior_variance, posterior_log_variance_clipped
    
    def p_mean_variance(self, x, cond, t, returns=None):
        '''
            p(x_t | x_{t-1}, c)
        '''
        if self.model.calc_energy:
            assert self.predict_epsilon
            x = torch.tensor(x, requires_grad=True)
            t = torch.tensor(t, dtype=torch.float, requires_grad=True)
            returns = torch.tensor(returns, requires_grad=True)
        
        if self.returns_condition:
            # epsilon could be epsilon or x0 itself
            epsilon_cond = self.model(x, cond, t, returns, use_dropout=False)
            epsilon_uncond = self.model(x, cond, t, returns, force_dropout=True)
            epsilon = epsilon_uncond + self.condition_guidance_w*(epsilon_cond - epsilon_uncond)
        else:
            epsilon = self.model(x, cond, t)
        
        t = t.detach().to(torch.int64)
        x_recon = self.predict_start_from_noise(x, t=t, noise=epsilon)

        if self.clip_denoised:
            x_recon.clamp_(-1., 1.)
        else:
            assert RuntimeError()
        
        model_mean, posterior_variance, posterior_log_variance = self.q_posterior(
                x_start=x_recon, x_t=x, t=t)
        
        return model_mean, posterior_variance, posterior_log_variance
    
    @torch.no_grad()
    def p_sample(self, x, cond, t, returns=None):
        '''
            p(x_t | x_{t-1}, c)
        '''
        b, *_, device = *x.shape, x.device
        model_mean, _, model_log_variance = self.p_mean_variance(x=x, cond=cond, t=t, returns=returns)
        noise = 0.5*torch.randn_like(x)
        # no noise when t == 0
        nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
        return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
    
    @torch.no_grad()
    def p_sample_loop(self, shape, cond, returns=None, verbose=True, return_diffusion=False):
        '''
            generate samples from prior
        '''
        device = self.betas.device
        
        batch_size = shape[0]
        x = 0.5*torch.randn(shape, device=device)
        x = apply_conditioning(x, cond, self.action_dim)

        if return_diffusion: diffusion = [x]

        # progress = utils.Progress(self.n_timesteps) if verbose else utils.Silent()
        for i in reversed(range(0, self.n_timesteps)):
            timesteps = torch.full((batch_size,), i, device=device, dtype=torch.long)
            x = self.p_sample(x, cond, timesteps, returns)
            x = apply_conditioning(x, cond, self.action_dim)

            # progress.update({'t': i})

            if return_diffusion: diffusion.append(x)
        
        if return_diffusion:
            return x, torch.stack(diffusion, dim=1)
        else:
            return x
    
    @torch.no_grad()
    def conditional_sample(self, cond, returns=None, horizon=None, *args, **kwargs):
        '''
            conditions : [ (time, state), ... ]
        '''
        device = self.betas.device
        batch_size = len(cond[0])
        horizon = horizon or self.horizon
        shape = (batch_size, horizon, self.transition_dim)

        return self.p_sample_loop(shape, cond, returns, *args, **kwargs)
    
    def grad_p_sample(self, x, cond, t, returns=None):
        b, *_, device = *x.shape, x.device
        model_mean, _, model_log_variance = self.p_mean_variance(x=x, cond=cond, t=t, returns=returns)
        noise = 0.5*torch.randn_like(x)
        # no noise when t == 0
        nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
        return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
    
    def grad_p_sample_loop(self, shape, cond, returns=None, verbose=True, return_diffusion=False):
        device = self.betas.device

        batch_size = shape[0]
        x = 0.5*torch.randn(shape, device=device)
        x = apply_conditioning(x, cond, self.action_dim)

        if return_diffusion: diffusion = [x]

        # progress = utils.Progress(self.n_timesteps) if verbose else utils.Silent()
        for i in reversed(range(0, self.n_timesteps)):
            timesteps = torch.full((batch_size,), i, device=device, dtype=torch.long)
            x = self.grad_p_sample(x, cond, timesteps, returns)
            x = apply_conditioning(x, cond, self.action_dim)

            # progress.update({'t': i})

            if return_diffusion: diffusion.append(x)

        # progress.close()

        if return_diffusion:
            return x, torch.stack(diffusion, dim=1)
        else:
            return x
    
    def grad_conditional_sample(self, cond, returns=None, horizon=None, *args, **kwargs):
        '''
            conditions : [ (time, state), ... ]
        '''
        device = self.betas.device
        batch_size = len(cond[0])
        horizon = horizon or self.horizon
        shape = (batch_size, horizon, self.transition_dim)

        return self.grad_p_sample_loop(shape, cond, returns, *args, **kwargs)

    #------------------------------------------ training ------------------------------------------#
    def q_sample(self, x_start, t, noise=None):
        if noise is None:
            noise = torch.randn_like(x_start)

        sample = (
            extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
            extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
        )

        return sample
    
    def p_losses(self, x_start, cond, t, returns=None):
        '''
            loss for p(x_t | x_{t-1}, c)
        '''
        noise = torch.randn_like(x_start)

        if self.predict_epsilon:
            # Cause we condition on obs at t=0
            noise[:, 0, self.action_dim:] = 0

        x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
        x_noisy = apply_conditioning(x_noisy, cond, self.action_dim)

        if self.model.calc_energy:
            assert self.predict_epsilon
            x_noisy.requires_grad = True
            t = torch.tensor(t, dtype=torch.float, requires_grad=True)
            returns.requires_grad = True
            noise.requires_grad = True
        
        x_recon = self.model(x_noisy, cond, t, returns)

        if not self.predict_epsilon:
            x_recon = apply_conditioning(x_recon, cond, self.action_dim)

        assert noise.shape == x_recon.shape

        if self.predict_epsilon:
            loss, info = self.loss_fn(x_recon, noise)
        else:
            loss, info = self.loss_fn(x_recon, x_start)

        return loss, info
    
    def loss(self, x, cond, returns=None):
        batch_size = len(x)
        t = torch.randint(0, self.n_timesteps, (batch_size,), device=x.device).long()
        return self.p_losses(x, cond, t, returns)
    
    def forward(self, cond, *args, **kwargs):
        return self.conditional_sample(cond=cond, *args, **kwargs)
    





class GaussianInvDynDiffusion(nn.Module):
    def __init__(self, model, horizon, observation_dim, action_dim, n_timesteps=200,
        loss_type='l1', clip_denoised=False, predict_epsilon=True, hidden_dim=256,
        action_weight=1.0, loss_discount=1.0, loss_weights=None, returns_condition=True,
        condition_guidance_w=0.1, ar_inv=True, train_only_inv=False):
        super().__init__()
    
        self.horizon = horizon
        self.observation_dim = observation_dim
        self.action_dim = action_dim
        self.transition_dim = observation_dim + action_dim
        self.model = model
        self.ar_inv = ar_inv
        self.train_only_inv = train_only_inv
        if self.ar_inv:
            self.inv_model = ARInvModel(hidden_dim=hidden_dim, observation_dim=observation_dim, action_dim=action_dim).to(device='cuda')
        else:
            self.inv_model = nn.Sequential(
                nn.Linear(2 * self.observation_dim, hidden_dim),
                nn.ReLU(),
                nn.Linear(hidden_dim, hidden_dim),
                nn.ReLU(),
                nn.Linear(hidden_dim, self.action_dim),
            ).to(device='cuda')
        self.returns_condition = returns_condition
        self.condition_guidance_w = condition_guidance_w

        betas = cosine_beta_schedule(n_timesteps)
        alphas = 1. - betas
        alphas_cumprod = torch.cumprod(alphas, axis=0).to(device='cuda')
        alphas_cumprod_prev = torch.cat([torch.ones(1).to(device='cuda'), alphas_cumprod[:-1]])

        self.n_timesteps = int(n_timesteps)
        self.clip_denoised = clip_denoised
        self.predict_epsilon = predict_epsilon

        self.register_buffer('betas', betas)
        self.register_buffer('alphas_cumprod', alphas_cumprod)
        self.register_buffer('alphas_cumprod_prev', alphas_cumprod_prev)

        # calculations for diffusion q(x_t | x_{t-1}) and others
        self.register_buffer('sqrt_alphas_cumprod', torch.sqrt(alphas_cumprod).to(device='cuda'))
        self.register_buffer('sqrt_one_minus_alphas_cumprod', torch.sqrt(1. - alphas_cumprod).to(device='cuda'))
        self.register_buffer('log_one_minus_alphas_cumprod', torch.log(1. - alphas_cumprod).to(device='cuda'))
        self.register_buffer('sqrt_recip_alphas_cumprod', torch.sqrt(1. / alphas_cumprod).to(device='cuda'))
        self.register_buffer('sqrt_recipm1_alphas_cumprod', torch.sqrt(1. / alphas_cumprod - 1).to(device='cuda'))

        # calculations for posterior q(x_{t-1} | x_t, x_0)
        posterior_variance = betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod)
        self.register_buffer('posterior_variance', posterior_variance)

        ## log calculation clipped because the posterior variance
        ## is 0 at the beginning of the diffusion chain
        self.register_buffer('posterior_log_variance_clipped',
            torch.log(torch.clamp(posterior_variance, min=1e-20)))
        self.register_buffer('posterior_mean_coef1',
            betas * torch.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod))
        self.register_buffer('posterior_mean_coef2',
            (1. - alphas_cumprod_prev) * torch.sqrt(alphas) / (1. - alphas_cumprod))

        ## get loss coefficients and initialize objective
        loss_weights = self.get_loss_weights(loss_discount)
        self.loss_fn = Losses['state_l2'](loss_weights)

    def get_loss_weights(self, discount):
        '''
            sets loss coefficients for trajectory

            action_weight   : float
                coefficient on first action loss
            discount   : float
                multiplies t^th timestep of trajectory loss by discount**t
            weights_dict    : dict
                { i: c } multiplies dimension i of observation loss by c
        '''
        self.action_weight = 1
        dim_weights = torch.ones(self.observation_dim, dtype=torch.float32).to(device='cuda')

        ## decay loss with trajectory timestep: discount**t
        discounts = discount ** torch.arange(self.horizon, dtype=torch.float).to(device='cuda')
        discounts = discounts / discounts.mean()
        loss_weights = torch.einsum('h,t->ht', discounts, dim_weights)
        # Cause things are conditioned on t=0
        if self.predict_epsilon:
            loss_weights[0, :] = 0

        return loss_weights
    
     #------------------------------------------ sampling ------------------------------------------#

    def predict_start_from_noise(self, x_t, t, noise):
        '''
            if self.predict_epsilon, model output is (scaled) noise;
            otherwise, model predicts x0 directly
        '''
        if self.predict_epsilon:
            return (
                extract(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
                extract(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
            )
        else:
            return noise
    
    def q_posterior(self, x_start, x_t, t):
        posterior_mean = (
            extract(self.posterior_mean_coef1, t, x_t.shape) * x_start +
            extract(self.posterior_mean_coef2, t, x_t.shape) * x_t
        )
        posterior_variance = extract(self.posterior_variance, t, x_t.shape)
        posterior_log_variance_clipped = extract(self.posterior_log_variance_clipped, t, x_t.shape)
        return posterior_mean, posterior_variance, posterior_log_variance_clipped

    def p_mean_variance(self, x, cond, t, returns=None):
        # print(returns )
        if self.returns_condition:
            # epsilon could be epsilon or x0 itself
            epsilon_cond = self.model(x, cond, t, returns, use_dropout=False)
            epsilon_uncond = self.model(x, cond, t, returns, force_dropout=True)
            epsilon = epsilon_uncond + self.condition_guidance_w*(epsilon_cond - epsilon_uncond)
        else:
            epsilon = self.model(x, cond, t)

        t = t.detach().to(torch.int64)
        x_recon = self.predict_start_from_noise(x, t=t, noise=epsilon)

        if self.clip_denoised:
            x_recon.clamp_(-1., 1.)
        else:
            assert RuntimeError()

        model_mean, posterior_variance, posterior_log_variance = self.q_posterior(
                x_start=x_recon, x_t=x, t=t)
        return model_mean, posterior_variance, posterior_log_variance
    
    @torch.no_grad()
    def p_sample(self, x, cond, t, returns=None):
        # print(returns)
        b, *_, device = *x.shape, x.device
        model_mean, _, model_log_variance = self.p_mean_variance(x=x, cond=cond, t=t, returns=returns)
        noise = 0.5*torch.randn_like(x)
        # no noise when t == 0
        nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
        return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise

    @torch.no_grad()
    def p_sample_loop(self, shape, cond, returns=None, verbose=True, return_diffusion=False):
        device = self.betas.device

        batch_size = shape[0]
        x = 0.5*torch.randn(shape, device=device)
        x = apply_conditioning(x, cond, 0)

        if return_diffusion: diffusion = [x]

        # progress = utils.Progress(self.n_timesteps) if verbose else utils.Silent()
        # print(returns)
        for i in reversed(range(0, self.n_timesteps)):
            timesteps = torch.full((batch_size,), i, device=device, dtype=torch.long)
            x = self.p_sample(x, cond, timesteps, returns)
            x = apply_conditioning(x, cond, 0)

            # progress.update({'t': i})

            if return_diffusion: diffusion.append(x)

        # progress.close()

        if return_diffusion:
            return x, torch.stack(diffusion, dim=1)
        else:
            return x

    @torch.no_grad()
    def conditional_sample(self, cond, returns=None, horizon=None, *args, **kwargs):
        '''
            conditions : [ (time, state), ... ]
        '''
        # print(returns)
        device = self.betas.device
        batch_size = len(cond[0])
        horizon = horizon or self.horizon
        shape = (batch_size, horizon, self.observation_dim)

        return self.p_sample_loop(shape, cond, returns, *args, **kwargs)


    #------------------------------------------ training ------------------------------------------#

    def q_sample(self, x_start, t, noise=None):
        if noise is None:
            noise = torch.randn_like(x_start)

        sample = (
            extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
            extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
        )

        return sample

    def p_losses(self, x_start, cond, t, returns=None):
        noise = torch.randn_like(x_start)

        x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
        x_noisy = apply_conditioning(x_noisy, cond, 0)

        x_recon = self.model(x_noisy, cond, t, returns)

        if not self.predict_epsilon:
            x_recon = apply_conditioning(x_recon, cond, 0)

        assert noise.shape == x_recon.shape

        if self.predict_epsilon:
            loss, info = self.loss_fn(x_recon, noise)
        else:
            loss, info = self.loss_fn(x_recon, x_start)

        return loss, info

    def loss(self, x, cond, returns=None):
        # print(x)
        if self.train_only_inv:
            # Calculating inv loss
            x_t = x[:, :-1, self.action_dim:]
            a_t = x[:, :-1, :self.action_dim]
            x_t_1 = x[:, 1:, self.action_dim:]
            x_comb_t = torch.cat([x_t, x_t_1], dim=-1)
            x_comb_t = x_comb_t.reshape(-1, 2 * self.observation_dim)
            a_t = a_t.reshape(-1, self.action_dim)
            if self.ar_inv:
                loss = self.inv_model.calc_loss(x_comb_t, a_t)
                info = {'a0_loss':loss}
            else:
                pred_a_t = self.inv_model(x_comb_t)
                loss = F.mse_loss(pred_a_t, a_t)
                info = {'a0_loss': loss}
        else:
            batch_size = len(x)
            # print(x.device)
            t = torch.randint(0, self.n_timesteps, (batch_size,), device=x.device).long()
            diffuse_loss, info = self.p_losses(x[:, :, self.action_dim:], cond, t, returns)
            # Calculating inv loss
            x_t = x[:, :-1, self.action_dim:]
            a_t = x[:, :-1, :self.action_dim]
            x_t_1 = x[:, 1:, self.action_dim:]
            x_comb_t = torch.cat([x_t, x_t_1], dim=-1)
            x_comb_t = x_comb_t.reshape(-1, 2 * self.observation_dim)
            a_t = a_t.reshape(-1, self.action_dim)
            if self.ar_inv:
                inv_loss = self.inv_model.calc_loss(x_comb_t, a_t)
            else:
                pred_a_t = self.inv_model(x_comb_t)
                inv_loss = F.mse_loss(pred_a_t, a_t)

            loss = (1 / 2) * (diffuse_loss + inv_loss)

        return loss, info

    def forward(self, cond, *args, **kwargs):
        return self.conditional_sample(cond=cond, *args, **kwargs)






















    
class ARInvModel(nn.Module):
    def __init__(self, hidden_dim, observation_dim, action_dim, low_act=-1.0, up_act=1.0):
        super(ARInvModel, self).__init__()
        self.observation_dim = observation_dim
        self.action_dim = action_dim

        self.action_embed_hid = 128
        self.out_lin = 128
        self.num_bins = 80

        self.up_act = up_act
        self.low_act = low_act
        self.bin_size = (self.up_act - self.low_act) / self.num_bins
        self.ce_loss = nn.CrossEntropyLoss()

        self.state_embed = nn.Sequential(
            nn.Linear(2 * self.observation_dim, hidden_dim),
            nn.ReLU(),
            nn.Linear(hidden_dim, hidden_dim),
            nn.ReLU(),
            nn.Linear(hidden_dim, hidden_dim),
            nn.ReLU(),
            nn.Linear(hidden_dim, hidden_dim),
        )

        self.lin_mod = nn.ModuleList([nn.Linear(i, self.out_lin) for i in range(1, self.action_dim)])
        self.act_mod = nn.ModuleList([nn.Sequential(nn.Linear(hidden_dim, self.action_embed_hid), nn.ReLU(),
                                                    nn.Linear(self.action_embed_hid, self.num_bins))])

        for _ in range(1, self.action_dim):
            self.act_mod.append(
                nn.Sequential(nn.Linear(hidden_dim + self.out_lin, self.action_embed_hid), nn.ReLU(),
                              nn.Linear(self.action_embed_hid, self.num_bins)))

    def forward(self, comb_state, deterministic=False):
        state_inp = comb_state

        state_d = self.state_embed(state_inp)
        lp_0 = self.act_mod[0](state_d)
        l_0 = torch.distributions.Categorical(logits=lp_0).sample()

        if deterministic:
            a_0 = self.low_act + (l_0 + 0.5) * self.bin_size
        else:
            a_0 = torch.distributions.Uniform(self.low_act + l_0 * self.bin_size,
                                              self.low_act + (l_0 + 1) * self.bin_size).sample()

        a = [a_0.unsqueeze(1)]

        for i in range(1, self.action_dim):
            lp_i = self.act_mod[i](torch.cat([state_d, self.lin_mod[i - 1](torch.cat(a, dim=1))], dim=1))
            l_i = torch.distributions.Categorical(logits=lp_i).sample()

            if deterministic:
                a_i = self.low_act + (l_i + 0.5) * self.bin_size
            else:
                a_i = torch.distributions.Uniform(self.low_act + l_i * self.bin_size,
                                                  self.low_act + (l_i + 1) * self.bin_size).sample()

            a.append(a_i.unsqueeze(1))
        # print(a)
        # a = torch.softmax(a)
        # print(a)
        a = torch.cat(a, dim=1)
        # print(a)
        # a = torch.softmax(a,dim=1)
        # print(a)
        return a
    
    def calc_loss(self, comb_state, action):
        eps = 1e-8
        action = torch.clamp(action, min=self.low_act + eps, max=self.up_act - eps)
        l_action = torch.div((action - self.low_act), self.bin_size, rounding_mode='floor').long()
        state_inp = comb_state

        state_d = self.state_embed(state_inp)
        loss = self.ce_loss(self.act_mod[0](state_d), l_action[:, 0])

        for i in range(1, self.action_dim):
            loss += self.ce_loss(self.act_mod[i](torch.cat([state_d, self.lin_mod[i - 1](action[:, :i])], dim=1)),
                                     l_action[:, i])

        return loss/self.action_dim