jxx123 / firets Goto Github PK

HI developer,

Thanks for developing this package.

May I ask how to read the training model details? For example, I use XGboost model as the estimator and put into the NARX model. After that I would like to use SHAP get the related explanations. Can I do like this?

BTW, I would like to confirm the meaning of exog_order, for example, exog_order = np.tile(10,6)). Does the 10 represent the the lag order of the exogenous variable and the 6 represent that "there are 6 exogenous variables", although there are 6 exogenous variables in the X_train?

getting following error

ValueError: The number of columns of X must be the same as the length of exog_order.

my X has 7 column and Y has one column. how I run this.

Hi.
I want to create a NARX (Nonlinear Autoregressive with exogenous variables) model based on LM (Levenberg Marquardt) method.

Since this two method are not implemented in keras, I search for the library fireTs (for NARX) and neupy (for LM).

I'm using the sample code for both libraries:

fireTS (NARX): fireTS
neupy (LM): Neupy

and I combine them:

from fireTS.models import NARX
from sklearn.ensemble import RandomForestRegressor
import numpy as np
from neupy import algorithms
from neupy.layers import *

x = np.array([[1, 2], [3, 4]])
y = np.array([[1], [0]])

#y = np.ravel(y) <-just to avoid a warning (the error is the same without comment)

network = Input(2) >> Sigmoid(3) >> Sigmoid(1)
optimizer = algorithms.LevenbergMarquardt(network)

mdl1 = NARX(
    optimizer, #I change random forest for LevenbergMarquardt
    auto_order=2,
    exog_order=[2, 2],
    exog_delay=[1, 1])

mdl1.fit(x, y)

ypred1 = mdl1.predict(x, y)

ypred1

But I'm having this error in .fit method:

ZeroDivisionError Traceback (most recent call last)

in ()
18 exog_delay=[1, 1])
19
---> 20 mdl1.fit(x, y)
21
22 ypred1 = mdl1.predict(x, y)

/usr/local/lib/python3.7/dist-packages/neupy/utils/iters.py in
count_minibatches(inputs, batch_size)
22
23 def count_minibatches(inputs, batch_size):
---> 24 return int(math.ceil(count_samples(inputs) / batch_size))
25
26

ZeroDivisionError: division by zero

Any solution?

Hi,
I am just wondering after playing around with the package using my own dataset. I split the data into training and the last 180 data as testing dataset. The fitted model gives good result compared to testing data but when applying predict procedure back to the training data the result is terrible. Is it normal? or am I doing a mistake?
Please advice,

Thank you

Hi, the package looks very fantastic to predict time series data. I was wondering if it is possible to implement it in an online training.

Hi Jinyu,

Regarding the update in OutputLagFeatureProcessor, perhaps a little off-topic, but I wonder what would be more memory efficient other than deque here?

Thanks!

Hi,
I want to access the paper mentioned on this repository, but it seems the link is broken.

Regards

Line 135 in the forecast method of models.py has this: deque(X[(-1 - d):(-1 - q):-1, i])

I believe the middle slice is missing a term; it should be deque(X[(-1 - d):(-1 - d - q):-1, i]), otherwise if the exogenous regressor(s) are given a delay that is >= the exog_order, this will return an empty deque and the forecast method will fail.

when iam trying to fit LSTM in NARX architecture it showing error like 'lstm does not have set_parameters'
brother could you please add some example code to fit LSTM in your model

Thank you for sharing this code!

Apologies if I'm misunderstanding something, but in the "Basic usage" and "Grid Search" example notebooks, it appears that the CGM column has been defined as the target variable as well as one of the two features in the design matrix. This is true for both the training and test sets. Having CGM as a predictor for CGM will lead to very low error. :)

I would have fixed this myself, but in the "Basic Usage" notebook, the markdown says that the design matrix is supposed to be insulin and meal data as the exogenous inputs. It wasn't obvious which of the columns was supposed to be the meal data, and I couldn't get any clarity by looking at the code for simglucose (which looks awesome, btw).

In File "/root/miniconda3/envs/myconda/lib/python3.8/site-packages/fireTS/models.py", line 3.
from sklearn.metrics.regression import r2_score, mean_squared_error

It raised a moduleNotFoundError: No module named 'sklearn.metrics.regression'.

The enviroment is: Ubuntu18.04, python3.8.12, scikit-learn 1.0.2; numpy 1.19.5; scipy 1.8.0

Implement a forecast method that is doing the following:

Given X(0), ... , X(t) and Y(0), ..., Y(t), output the forecast values of Y as a vector Y_hat(t+1), Y_hat(t+2), ... ,Y_hat(t+k).

Hi Jinyu,

When we are plotting the prediction results when using DirectAutoRegressor, should we be only setting the first pred_step elements to be np.nan?

EDIT: I meant to say should we be plotting the results starting from index pred_step.

I was also wondering why you set the first pred_step + max(auto_order-1, max(exog_order+exog_delay)-1) elements to be np.nan.

Thanks!
Daniel

Why the forecast method need the target value?
mdl1.predict(x, y, step=3)

I know that there is a way for training a MIMO system by training multiple MISO systems and get multiple models. But is there any way to get only a single model?