What happened + What you expected to happen

I tried a ppo agent with these configs: model_config['batch_mode'] = 'complete_episodes' , model_config['use_gae'] = 'False',model_config['use_critic'] = 'False' then I got this error:" /usr/local/lib/python3.7/dist-packages/ray/rllib/policy/torch_policy.py in (s) 701 data.append( 702 tree.map_structure( --> 703 lambda s: s.to(self.device), tower.tower_stats[stats_name] 704 ) 705 )

AttributeError: 'float' object has no attribute 'to'" so to be sure I wrote a dummy env and tested it again and I got same error. why I'am doing this config? well, I've manipulated postprocess so I can't go truncated episode and thus I can't use gae.

Versions / Dependencies

ray 1.12.1 google colab python 3.7.13

Reproduction script

import gym from gym import spaces import numpy as np import torch from torch import nn from ray.rllib.models.torch.torch_modelv2 import TorchModelV2 from ray.rllib.agents.ppo import ppo import ray from ray.rllib.models import ModelCatalog class dummyvisoinenv(gym.Env): def init(self,config): self.length = length self.width = width self.channels = channels self.action_space = spaces.Tuple([spaces.Discrete(4),spaces.Discrete(4),spaces.Discrete(3)]) self.observation_space = spaces.Box(low=-1 * 10 10, high=10 10, shape=(self.length,self.width,self.channels), dtype=np.float32) self.episode_len = episode_len
def reset(self): self.timestamp = 0 self.done = False observation = np.random.rand(self.length,self.width,self.channels) return observation def step(self,action): a1, a2, a3 = action self.timestamp = self.timestamp + 1 if self.timestamp == self.episode_len: self.done = True reward = np.random.randint(0,5) observation = np.random.rand(self.length,self.width,self.channels) info = {} print(self.timestamp) return observation, reward, self.done, info

resnet model based on pytorch official

def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1): """3x3 convolution with padding""" return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=dilation, groups=groups, bias=False, dilation=dilation)

def conv1x1(in_planes, out_planes, stride=1): """1x1 convolution""" return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)

import torch

from torch import nn

from ray.rllib.models.torch.torch_modelv2 import TorchModelV2

class BasicBlock(nn.Module): expansion = 1

def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
             base_width=64, dilation=1, norm_layer=None):
    # from torch import nn
    nn.Module.__init__(self)
    if norm_layer is None:
        norm_layer = nn.BatchNorm2d
    if groups != 1 or base_width != 64:
        raise ValueError('BasicBlock only supports groups=1 and base_width=64')
    if dilation > 1:
        raise NotImplementedError("Dilation > 1 not supported in BasicBlock")
    # Both self.conv1 and self.downsample layers downsample the input when stride != 1
    self.conv1 = conv3x3(inplanes, planes, stride)
    self.bn1 = norm_layer(planes)
    self.relu = nn.ReLU(inplace=True)
    self.conv2 = conv3x3(planes, planes)
    self.bn2 = norm_layer(planes)
    self.downsample = downsample
    self.stride = stride

def forward(self, x):
    identity = x

    out = self.conv1(x)
    out = self.bn1(out)
    out = self.relu(out)

    out = self.conv2(out)
    out = self.bn2(out)

    if self.downsample is not None:
        identity = self.downsample(x)

    out += identity
    out = self.relu(out)

    return out

class Bottleneck(nn.Module):

Bottleneck in torchvision places the stride for downsampling at 3x3 convolution(self.conv2)

# while original implementation places the stride at the first 1x1 convolution(self.conv1)
# according to "Deep residual learning for image recognition"https://arxiv.org/abs/1512.03385.
# This variant is also known as ResNet V1.5 and improves accuracy according to
# https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch.

expansion = 4

def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
             base_width=64, dilation=1, norm_layer=None):
    super(Bottleneck, self).__init__()
    if norm_layer is None:
        norm_layer = nn.BatchNorm2d
    width = int(planes * (base_width / 64.)) * groups
    # Both self.conv2 and self.downsample layers downsample the input when stride != 1
    self.conv1 = conv1x1(inplanes, width)
    self.bn1 = norm_layer(width)
    self.conv2 = conv3x3(width, width, stride, groups, dilation)
    self.bn2 = norm_layer(width)
    self.conv3 = conv1x1(width, planes * self.expansion)
    self.bn3 = norm_layer(planes * self.expansion)
    self.relu = nn.ReLU(inplace=True)
    self.downsample = downsample
    self.stride = stride

def forward(self, x):
    identity = x

    out = self.conv1(x)
    out = self.bn1(out)
    out = self.relu(out)

    out = self.conv2(out)
    out = self.bn2(out)
    out = self.relu(out)

    out = self.conv3(out)
    out = self.bn3(out)

    if self.downsample is not None:
        identity = self.downsample(x)

    out += identity
    out = self.relu(out)

    return out

class ResNet(TorchModelV2,nn.Module):

def __init__(self, obs_space, action_space, num_outputs, model_config,
             name,block , layers = [2, 2, 2, 2], zero_init_residual=False,
             groups=1, width_per_group=64, replace_stride_with_dilation=None,
             norm_layer=None):
    TorchModelV2.__init__(self, obs_space, action_space, num_outputs,
            model_config, name)
    nn.Module.__init__(self)
    if norm_layer is None:
        norm_layer = nn.BatchNorm2d
    self._norm_layer = norm_layer

    self.inplanes = 64
    self.dilation = 1
    if replace_stride_with_dilation is None:
        # each element in the tuple indicates if we should replace
        # the 2x2 stride with a dilated convolution instead
        replace_stride_with_dilation = [False, False, False]
    if len(replace_stride_with_dilation) != 3:
        raise ValueError("replace_stride_with_dilation should be None "
                         "or a 3-element tuple, got {}".format(replace_stride_with_dilation))
    self.groups = groups
    self.base_width = width_per_group
    self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3,
                           bias=False)
    self.bn1 = norm_layer(self.inplanes)
    self.relu = nn.ReLU(inplace=True)
    self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
    self.layer1 = self._make_layer(block, 64, layers[0])
    self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
                                   dilate=replace_stride_with_dilation[0])
    self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
                                   dilate=replace_stride_with_dilation[1])
    self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
                                   dilate=replace_stride_with_dilation[2])
    self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
    self.fc = nn.Linear(512 * block.expansion, num_outputs)
    self.vfc = nn.Linear(512 * block.expansion, 1)

    for m in self.modules():
        if isinstance(m, nn.Conv2d):
            nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
        elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
            nn.init.constant_(m.weight, 1)
            nn.init.constant_(m.bias, 0)

    # Zero-initialize the last BN in each residual branch,
    # so that the residual branch starts with zeros, and each residual block behaves like an identity.
    # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
    if zero_init_residual:
        for m in self.modules():
            if isinstance(m, Bottleneck):
                nn.init.constant_(m.bn3.weight, 0)
            elif isinstance(m, BasicBlock):
                nn.init.constant_(m.bn2.weight, 0)

def _make_layer(self, block, planes, blocks, stride=1, dilate=False):
    norm_layer = self._norm_layer
    downsample = None
    previous_dilation = self.dilation
    if dilate:
        self.dilation *= stride
        stride = 1
    if stride != 1 or self.inplanes != planes * block.expansion:
        downsample = nn.Sequential(
            conv1x1(self.inplanes, planes * block.expansion, stride),
            norm_layer(planes * block.expansion),
        )

    layers = []
    layers.append(block(self.inplanes, planes, stride, downsample, self.groups,
                        self.base_width, previous_dilation, norm_layer))
    self.inplanes = planes * block.expansion
    for _ in range(1, blocks):
        layers.append(block(self.inplanes, planes, groups=self.groups,
                            base_width=self.base_width, dilation=self.dilation,
                            norm_layer=norm_layer))

    return nn.Sequential(*layers)

def _forward_impl(self, x):
    # print(x)
    # print(x.size())
    x = torch.permute(x,(0,3,1,2))
    # See note [TorchScript super()]
    x = self.conv1(x)
    x = self.bn1(x)
    x = self.relu(x)
    x = self.maxpool(x)

    x = self.layer1(x)
    x = self.layer2(x)
    x = self.layer3(x)
    x = self.layer4(x)

    x = self.avgpool(x)
    last_x = torch.flatten(x, 1)
    x = self.fc(last_x)
    self.value = self.vfc(last_x)

    return x

def forward(
    self,
    input ,
    state,
    seq_lens):
    x = input["obs"].float()
    return (self._forward_impl(x), state)

def value_function(self):
    assert self.value is not None, "must call forward() first"
    # print("1stvalue",self.value)
    self.value = torch.reshape(self.value , [-1])
    # print('secondvalue',self.value)
    return self.value

configs

length = 20 width = 20 channels = 3 episode_len = 10 env_config = {"length" : length , 'width': width , 'channels': channels,'episode_len' : episode_len} env_instance = dummyvisoinenv(env_config) model = ppo model_config = model.DEFAULT_CONFIG.copy() model_config["env"] = env_instance model_config['num_workers'] = 0 model_config['train_batch_size'] = 5 model_config["framework"]="torch" model_config['sgd_minibatch_size'] = 5 model_config["model"] = { "custom_model": 'resnet18torch', 'custom_model_config': { 'block':BasicBlock , 'layers':[2, 2, 2, 2], 'zero_init_residual': False , 'groups' : 1, 'width_per_group' : 64 , 'replace_stride_with_dilation' : None , 'norm_layer' : None } } model_config['use_gae'] = False model_config["use_critic"] = False model_config['batch_mode'] = 'complete_episodes' # ray.shutdown() ray.init(ignore_reinit_error=True) ModelCatalog.register_custom_model('resnet18torch', ResNet) total_episodes = 100 trainer = model.PPOTrainer(env= dummyvisoinenv, config=model_config) for i in range(total_episodes): result = trainer.train() print('done: ',i)

Issue Severity

Medium: It is a significant difficulty but I can work around it.

ray-project / ray

[<Ray component: RLlib>] ppo error when not using critic #24907