compat.Pytorch2River

[batman7090]

hello all,
I am getting a TypeError: forward() got an unexpected keyword argument 'n_features' I am not getting how to pass this parameters into the build model function could anybody shed some light on this? I have attached the code below.

Thank you in advance.

from river import compat
from river import datasets
from river import evaluate
from river import metrics
from river import preprocessing
from torch import nn
from torch import optim
import torch
import pandas as pd
from river import stream
import itertools

_ = torch.manual_seed(0)
n_features=3

def build_torch_mlp_regressor(n_features):
  net = nn.Sequential(
      nn.Linear(n_features,45),
      nn.Linear(45,50),
      nn.Linear(50,50),
      nn.Linear(50,3)
  )
  return net
model = compat.PyTorch2RiverRegressor(build_fn=build_torch_mlp_regressor(n_features),
                                      loss_fn=nn.MSELoss,
                                      optimizer_fn=optim.Adam)

df = pd.read_csv('/content/drive/MyDrive/data/Final_data/one_lakh_more_accurate.csv')
train_size = 100


for xb in pd.read_csv('/content/drive/MyDrive/data/Final_data/soft_robotics_2_scaled.csv', chunksize=50):
  y = xb.loc[:,'X':'Z']
  x = xb.loc[:,'P_set_1':'P_set_3']
  model.predict_one(x)
  model.learn_one(x,y)

Error I got

T

ypeError Traceback (most recent call last)
in ()
10 y = xb.loc[:,'X':'Z']
11 x = xb.loc[:,'P_set_1':'P_set_3']
---> 12 model.predict_one(x)
13 model.learn_one(x,y)
14

2 frames
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
1108 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1109 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110 return forward_call(*input, **kwargs)
1111 # Do not call functions when jit is used
1112 full_backward_hooks, non_full_backward_hooks = [], []

TypeError: forward() got an unexpected keyword argument 'n_features'

[MaxHalford]

Basically build_fn expects to be passed a function, and you're actually calling the function. This should work:

model = compat.PyTorch2RiverRegressor(build_fn=build_torch_mlp_regressor
                                      loss_fn=nn.MSELoss,
                                      optimizer_fn=optim.Adam)

Why is this? Because in online learning you the user should never have to worry about the number of features in the data stream. Models should handle this for you.

Note that we'll be removing PyTorch compatibility in the next release. It will instead be located in river-extra.

[batman7090]

@MaxHalford no it is not working could you please share with me the example of how can iter over CSV file with train test loop?
can I use PyTorch2RiverRegressor for multi-target regression?

thank you in advance

[MaxHalford]

could you please share with me the example of how can iter over CSV file with train test loop?

Take a look at iter_csv as well as iter_pandas.

can I use PyTorch2RiverRegressor for multi-target regression?

No, it doesn't support that. But I encourage you to consider the multioutput module. It contains model wrappers, which you can wrap around any regression model, which includes PyTorch2RiverRegressor.

[batman7090]

@MaxHalford, thank you so much it is working now.