sklearn scale was not recognized within the bracket

[pinghongyeh]

Hi Gidon,

I am testing the ce_prediction scripts, but it failed in the following

for node_i in tqdm(np.arange(num_nodes)): # loop over all nodes
# a boolean vector to remove the diagonal as a feature
exclude_diag = np.ones(num_nodes, dtype=bool)
exclude_diag[node_i] = False

inputs = [x_ce[:,node_i,:], x_cm[:,node_i,exclude_diag], x_sc[:,node_i,exclude_diag]] # select a node
inputs = [scale(x) for x in inputs] # standardize the inputs (mean = 0, std = 1)

for input_i, (input_name, input_x, input_df) in enumerate(zip(input_names, inputs, dfs)): # loop over the three input types
    for ind_kf, (train_index, test_index) in enumerate(kf.split(input_x)): # the 5-fold iterator
        curr_lm = clone(lm)
        curr_lm.fit(input_x[train_index, :], y[train_index])
        y_pred_cv[input_i, test_index, node_i] = curr_lm.predict(input_x[test_index, :])
        
    # keep the pearson correlation coefficient between the real and predicted age    
    input_df.loc[node_i, 'correlation_coef'], _ = stats.pearsonr(y, y_pred_cv[input_i, :, node_i])

NameError: name 'scale' is not defined

I have run the
from matplotlib import pyplot as plt
from matplotlib.colors import rgb2hex
import numpy as np
import seaborn as sns
import pandas as pd
from scipy import stats
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.preprocessing import scale
from sklearn.model_selection import KFold
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import SGDRegressor
from sklearn.base import clone
from sklearn.metrics import explained_variance_score
from tqdm import tqdm

It seems that scale(x) does not run within the bracket []

Do you have any suggestion?
Thank you.
Ping

[DsgGidL]

Hi @pinghongyeh thanks for spotting the issue, here is a quick fix:

from sklearn.preprocessing import scale

inputs = [scale(x) for x in inputs] # standardize the inputs (mean = 0, std = 1)

[GidLev]

b97053c
fixed in ce_prediction.ipynb

[pinghongyeh]

Sorry for my novice.
I do not see any change.
The error still exist.

Input In [94], in (.0)
13 exclude_diag[node_i] = False
15 inputs = [x_ce[:,node_i,:], x_cm[:,node_i,exclude_diag], x_sc[:,node_i,exclude_diag]] # select a node
---> 16 inputs = [scale(x) for x in inputs] # standardize the inputs (mean = 0, std = 1)
18 for input_i, (input_name, input_x, input_df) in enumerate(zip(input_names, inputs, dfs)): # loop over the three input types
19 for ind_kf, (train_index, test_index) in enumerate(kf.split(input_x)): # the 5-fold iterator

NameError: name 'scale' is not defined

Thank you.

[DsgGidL]

The fix was in the imports section, please run this import line before the scaling operation:
from sklearn.preprocessing import scale

Let me know if it solved it.

[pinghongyeh]

The problem of "scale" was solved.
I have other questions, if you don't mind.
In the "ce_subjects_pipeline" scripts, I was not able to find the "NKI_200_schaefer_sc_train_consensus_mat.npy". How was it created?
In addition, in the session of the paper, "2.12. Predicting group-level functional from structural connectivity with deep learning", are the scripts regarding the deep learning algorithms described in this session available to the public?

Thank you.

[DsgGidL]

Hi,

The consensus matrix was generated using Betzel et al. (2018) distance-dependent consensus thresholds.
I don't think I'll manage to find the file you are referring to but you can generate it using this code:
https://github.com/GidLev/consensus-thresholding

Predicting group-level functional from structural connectivity with deep learning, here is the relevant code:
https://gist.github.com/GidLev/0a5bb6ddef7df1a49b430ec3957cdc9e

[pinghongyeh]

Hi Gidon,
Thank you for providing the codes.
For the distance matrix, the required input for the fcn_group_bins, how can it be created using the example data, e.g. NKI_200_schaefer_sc_matrices.npz?

Ping