agents: AttributeError: 'tuple' object has no attribute 'rank'

Trying out the most basic example on

• Windows 10
• Python 3.7
• tensorflow 2.1.0
• tf-agents 0.4.0

Error i get

Traceback (most recent call last):
  File "src\agent.py", line 58, in <module>
    action, _states = agent.policy.action(obs)
  File "C:\Users\andre\.virtualenvs\ZeusTrader\lib\site-packages\tf_agents\policies\tf_policy.py", line 279, in action
    step = action_fn(time_step=time_step, policy_state=policy_state, seed=seed)
  File "C:\Users\andre\.virtualenvs\ZeusTrader\lib\site-packages\tf_agents\utils\common.py", line 154, in with_check_resource_vars
    return fn(*fn_args, **fn_kwargs)
  File "C:\Users\andre\.virtualenvs\ZeusTrader\lib\site-packages\tf_agents\policies\random_tf_policy.py", line 89, in _action
    outer_dims = nest_utils.get_outer_shape(time_step, self._time_step_spec)
  File "C:\Users\andre\.virtualenvs\ZeusTrader\lib\site-packages\tf_agents\utils\nest_utils.py", line 394, in get_outer_shape
    nested_tensor, spec, num_outer_dims=num_outer_dims):
  File "C:\Users\andre\.virtualenvs\ZeusTrader\lib\site-packages\tf_agents\utils\nest_utils.py", line 97, in is_batched_nested_tensors
    if any(spec_shape.rank is None for spec_shape in spec_shapes):
  File "C:\Users\andre\.virtualenvs\ZeusTrader\lib\site-packages\tf_agents\utils\nest_utils.py", line 97, in <genexpr>
    if any(spec_shape.rank is None for spec_shape in spec_shapes):
AttributeError: 'tuple' object has no attribute 'rank'

Code i run

import tensorflow as tf
from collections import Counter, defaultdict
from tf_agents.networks import q_network
from tf_agents.utils import common
from tf_agents.agents.dqn import dqn_agent
from tf_agents.agents.random.random_agent import RandomAgent
from tf_agents.environments import suite_gym
from environment import StockExchangeEnv01

# tried with and without..error persists
# tf.compat.v1.enable_v2_behavior()

learning_rate = 0.0001
optimizer = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)

# tried both my own Environment and the basic "cartpole-v0"
train_env = StockExchangeEnv01()
env_name = 'CartPole-v0'
#train_env = suite_gym.load(env_name)

train_env.reset()
print(train_env.action_spec())
"""
# Neural Net of the Agent. This NN will get x (env) and spit out y (action).
q_net = q_network.QNetwork(
  train_env.observation_spec(),
  train_env.action_spec(),
  fc_layer_params=(100,))
print(train_env.action_spec())

#
agent = dqn_agent.DqnAgent(
  train_env.time_step_spec(),
  train_env.action_spec(),
  q_network=q_net,
  optimizer=optimizer)
"""

# tried both..dqn agent and random agent

agent = RandomAgent(
    train_env.time_step_spec(),
    train_env.action_spec()
)
agent.initialize()

obs = train_env.reset()
actions = Counter()
pnl = defaultdict(float)
total_rewards = 0.0

for i in range(300):
    #action, _states = model.predict(obs)
    action, _states = agent.policy.action(obs)
    obs, rewards, dones, info = train_env.step(action)
    actions[action[0].item()] += 1
    pnl[action[0].item()] += rewards
    total_rewards += rewards
    if dones:
        break

print('actions : {}'.format(actions))
print('rewards : {}'.format(total_rewards))

The code in tf agents gets the ‘shape’ from the action_spec, which is a tuple in my case. Then it tries to retrieve key “rank” from a tuple.

What am i missing?