app.py

import streamlit as st
import numpy as np
import matplotlib.pyplot as plt

st.title('Fitting simple models with JAX')
st.header('A quadratric regression example')

st.markdown('*\"Parametrised models are simply functions that depend on inputs and trainable parameters. There is no fundamental difference between the two, except that trainable parameters are shared across training samples whereas the input varies from sample to sample.\"* [(Yann LeCun, Deep learning course)](https://atcold.github.io/pytorch-Deep-Learning/en/week02/02-1/#Parametrised-models)')

st.latex(r'''h(\boldsymbol x, \boldsymbol w)= \sum_{k=1}^{K}\boldsymbol w_{k} \phi_{k}(\boldsymbol x)''')


# Sidebar inputs
number_of_observations = st.sidebar.slider('Number of observations', min_value=50, max_value=150, value=75)
noise_standard_deviation = st.sidebar.slider('Standard deviation of the noise', min_value = 0.0, max_value=2.0, value=0.25)
cost_function = st.sidebar.radio('What cost function you want to use for the fitting?', options=('RMSE-Loss', 'Huber-Loss'))

np.random.seed(2)

X = np.column_stack((np.ones(number_of_observations), 
                     np.random.random(number_of_observations)))      

w = np.array([3.0, -20.0, 32.0])  # coefficients                                    

X = np.column_stack((X, X[:,1] ** 2))   # add x**2 column
additional_noise = 8 * np.random.binomial(1, 0.03, size = number_of_observations)
y = np.dot(X, w) + noise_standard_deviation * np.random.randn(number_of_observations) \
        + additional_noise	


fig, ax = plt.subplots(dpi=320)
ax.set_xlim((0,1))
ax.set_ylim((-5,26))
ax.scatter(X[:,1], y, c='#e76254' ,edgecolors='firebrick')

st.pyplot(fig)
st.write(X[:5, :])