How to obtain a single prediction from trained model?

[Karl]

I'm trying to obtain a single prediction from a trained model for production.

# Load the vecnormalize object
vec_normalize = VecNormalize.load("vecnormalize.pkl", envs)

# Load the agent
loaded_model = A2C.load("model_multi_vec.zip")

# Pass vecnormalize object to the loaded agent
loaded_model.set_env(vec_normalize)

# # Get the window of data to be used as an observation
single_obs = gymdf.iloc[-10:].values

# # Normalize the observation using the vecnormalize object
single_obs = envs.normalize_obs(single_obs)

# Perform the prediction
action, _states = loaded_model.predict(single_obs)
action, _states

But I'm getting this error

vec_normalize.py:212, in VecNormalize.normalize_obs(self, obs)
    210             obs_[key] = self._normalize_obs(obs[key], self.obs_rms[key]).astype(np.float32)
    211     else:
--> 212         obs_ = self._normalize_obs(obs, self.obs_rms).astype(np.float32)
    213 return obs_

vec_normalize.py:188, in VecNormalize._normalize_obs(self, obs, obs_rms)
    181 def _normalize_obs(self, obs: np.ndarray, obs_rms: RunningMeanStd) -> np.ndarray:
    182     """
    183     Helper to normalize observation.
    184     :param obs:
    185     :param obs_rms: associated statistics
    186     :return: normalized observation
    187     """
--> 188     return np.clip((obs - obs_rms.mean) / np.sqrt(obs_rms.var + self.epsilon), -self.clip_obs, self.clip_obs)

ValueError: operands could not be broadcast together with shapes (10,6) (10,2)

[Karl]

So reading the source code I noticed that it doesn't actually use OHLCV, only Close by default. So needed to implement the other ones like this to get it working. Now it makes sense the shape values differed

def normalize_data(env):
    start = env.frame_bound[0] - env.window_size
    end = env.frame_bound[1]
    prices = env.df.loc[:, 'Close'].to_numpy()[start:end]
    
    signal_features = env.df.loc[:, ['Open', 'High', 'Low', 'Close']].to_numpy()[start:end]
    diff = np.insert(np.diff(prices), 0, 0)
    signal_features = np.column_stack((signal_features, diff))
    
    return prices, signal_features


class MyForexEnv(ForexEnv):
    _process_data = normalize_data