Ridge python quantile LINEAR regression by Deep neural networks

do it for all main Deep NN packages , but do not use hidden layers
keras for example point to start https://www.tensorflow.org/tutorials/keras/regression
tensor flow
fast.ai
pytorch

then it should be 4 solutions, all of them should provide good performance
each solution in separate file

data is mixture of categorical and continues features
at least 5 categorical features
at least 5 continues features
at least 100000 rows
data you find : several datasets ( more than 2)

see
primitive linear (or log-linear) model W \cdot \mathbf{x} + \mathbf{b} (where W is a weight matrix, \mathbf{x} = (x_1, x_2) is an input vector, and \mathbf{b} is a bias vector) without hidden layers,
https://tsmatz.wordpress.com/2017/09/13/overfitting-for-regression-and-deep-learning/

better to find and use existing code from web, then you include web links for source

to prove all done correctly: make prediction manually use only one dimension vector coefficients
like
https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.Ridge.html
coef_ : ndarray of shape (n_features,) or (n_targets, n_features)
Weight vector(s).
you prove correctness by upper prediction = Data * coef_upper_
you prove correctness by lower prediction = Data * coef_lower_
coef_upper_
coef_lower_
like
https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.QuantileRegressor.html
coef_ : array of shape (n_features,)
Estimated coefficients for the features

you baseline
https://github.com/msangnier/qreg
https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.QuantileRegressor.html
https://scikit-learn.org/stable/auto_examples/ensemble/plot_gradient_boosting_quantile.html