Hyper parameter tunning

[batman7090]

Hello @MaxHalford, @ALL My application is incremental multi-target regression(3input features and 3ouput targets), Using iSOUPTreeRegressor the model gets trained and works okay. Still, If I tweak the hyper parameter It won't show the difference in the output. How to tune the hyperparameter in this case?
I also tried using MLPregressor but the model doesn't try to learn. when I predict the output It gives zeros as output.

thank you in advance

[MaxHalford]

Hey! Not sure what you're asking here. Are you saying that changing a hyperparameter has no impact? If so, it's a bit hard for me to help if I don't see any code.

[batman7090]

Hello @MaxHalford, I am really excited that you replied. thank you so much, here is my code when I used MLPregressor.

from river import neural_net as nn
from river import optim
from river import preprocessing as pp
import numpy as np

model = (
    pp.StandardScaler()|
    nn.MLPRegressor(
        hidden_dims=(45,50),
        activations=(nn.activations.ReLU,
        nn.activations.ReLU,
        nn.activations.ReLU,
        nn.activations.Identity),
        optimizer=optim.SGD(1e-3),
        seed=42
    )
)

batch_size = 10000
n_targets = 3
n_samples = 0
max_samples = 6000000 
y_pred = np.zeros((max_samples, n_targets))
y_true = np.zeros((max_samples, n_targets))
counts = np.zeros((max_samples, 1))

for epoch in range(5):
  for xb in pd.read_csv('/content/drive/MyDrive/data/data_13_may_eve_changed.csv', chunksize=batch_size):
    y = xb.loc[:,'X':'Z']
    x = xb.loc[:,'P_set_1':'P_set_3']
    y_pred[n_samples] = model.predict_many(x).to_numpy()[0]
    model = model.learn_many(x,y)
    if n_samples >=
    n_samples+=batch_size
  print('Epoch ',epoch, ' is completed')

===========================
========Input to MLP===========

P_set_1	P_set_2	P_set_3	X	Y	Z
0	0.00000	0.00000	0.000000	0.000047	0.000097	0.135113
1	0.00000	0.00000	0.000000	0.000047	0.000097	0.135113
2	0.00000	0.00000	0.000000	0.000047	0.000097	0.135113
3	0.00000	0.00000	0.000000	0.000048	0.000096	0.135113
4	0.00000	0.00000	0.000000	0.000048	0.000096	0.135113
...	...	...	...	...	...	...
5961315	60472.39474	80586.20755	1441.397712	0.009286	0.036996	0.139082
5961316	60470.00217	80587.99201	1442.005816	0.009286	0.036996	0.139082
5961317	60467.60957	80589.77638	1442.614045	0.009282	0.036999	0.139081
5961318	60465.21694	80591.56066	1443.222401	0.009277	0.037001	0.139080
5961319	60462.82427	80593.34485	1443.830884	0.009277	0.037001	0.139080

here p_set_1,...,P_set_3 is the input features and output label is X,Y,Z are the co-ordinates this is amulti-target regression.
I used isoptreeregressor regressor but the output was not satisfying also I tried tunning hyperparameter manually still the results were no good or more error.

from skmultiflow.data import FileStream
from skmultiflow.trees import iSOUPTreeRegressor
import numpy as np
import sys

sys.setrecursionlimit(1000000)

n_targets = 3
stream = FileStream('/content/drive/MyDrive/data/Final_data/soft_robotics_2_scaled.csv', target_idx=3, n_targets=n_targets)


model = iSOUPTreeRegressor(leaf_prediction='adaptive')

n_samples = 0
max_samples = 90000
y_pred = np.zeros((max_samples, n_targets))
y_true = np.zeros((max_samples, n_targets))
counts = np.zeros((max_samples, 1))

while n_samples < max_samples and stream.has_more_samples():
    X, y = stream.next_sample()
    y_true[n_samples] = y[0]
    y_pred[n_samples] = model.predict(X)[0]
    counts[n_samples] = n_samples
    model.partial_fit(X, y)
    n_samples += 1

thank you in advance.

[batman7090]

sorry for the long reply, there was no impact on the output by changing the hyperparameter of the iSouptree regressor.

[MaxHalford]

Thanks! I'll take a look sometime this week 🤝

(I formatted your post to highlight the code; please do that yourself next time, as it makes everyone's lives a little bit easier)

[smastelini]

Hey @batman7090, from your code I see you're using skmultiflow for the tree regressor. You should use river instead, as skmultiflow is not under development anymore.