Vlad Niculae vene

## hyperbolic.py
"""wrapped hyperbolic distributions

following https://arxiv.org/abs/1902.02992

"""

# author: vlad niculae <v.niculae@uva.nl>
# license: bsd 3-clause

import torch

## check_det_without_rotat.py
""" trying to understand the determinant of sphere-to-cyl """

import numpy as np
import jax.numpy as jnp
from jax import jacfwd


# map from cylinder to sphere
def phi(zr):
    d = zr.shape[0]

## kl.py
"""
Approximating the cross-entropy between two Power Sphericals.

Uses a second-order Taylor expansion to approximate E[log(1+z)].
"""

# author: vlad n <vlad@vene.ro>
# license: mit
# documentation: https://hackmd.io/@vladn/SJ93wMevK

## plot_dual_norms.py
"""
Dual p-norms illustrated.

For any norm |.|, the dual norm is defined as |y|_* = max{ <x, y> for |x| <= 1 }.

The figure shows the unit balls of the p-norm, for p = 1.5, 2, and 3.

We compute the dual norm at a dual vector y (short black arrow), rotating
uniformly around the origin over time.

## multi_mixins.py
class Base:

    def say(self, val):
        print("base says", val)


class A(Base):

    def say(self, val):
        print("say A")

## README.txt
Example input and output.

$ python conf.py seed=42 lr=.1

project: ???
seed: 42
lr: 0.1
epochs: ???
p_drop: 0.5
baseconf: null

## decision_boundary.py
# author: vlad niculae <vlad@vene.ro>
# license: mit

import torch
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.colors as colors

from entmax import sparsemax, entmax15
from entmax.losses import sparsemax_loss, entmax15_loss

## soft_box.py
# author: vn

import numpy as np
from scipy.optimize import root_scalar
import torch
import matplotlib.pyplot as plt


def entropy(y, a, b):

## energy.py
# Density estimation with energy-based models
# Langevin sampling, contrastive divergence training.
# Author: Vlad Niculae <vlad@vene.ro>
# License: MIT

import numpy as np
import torch
from sklearn import datasets
import matplotlib.pyplot as plt

## check_spherical_jacob_tangent.py
# Geometric intepretation of the gradient of the mapping:
# f : (0, inf) x Sphere(k-1) -> R^k
# f(r, u) -> r*u
# The *catch*: R can vary on (0, inf) but u may only vary on the
# k-1--dimensional tangent plane!

import numpy as np

def main():
	"""wrapped hyperbolic distributions

	following https://arxiv.org/abs/1902.02992

	"""

	# author: vlad niculae <v.niculae@uva.nl>
	# license: bsd 3-clause

	import torch
	""" trying to understand the determinant of sphere-to-cyl """

	import numpy as np
	import jax.numpy as jnp
	from jax import jacfwd


	# map from cylinder to sphere
	def phi(zr):
	d = zr.shape[0]
	"""
	Approximating the cross-entropy between two Power Sphericals.

	Uses a second-order Taylor expansion to approximate E[log(1+z)].
	"""

	# author: vlad n <vlad@vene.ro>
	# license: mit
	# documentation: https://hackmd.io/@vladn/SJ93wMevK
	"""
	Dual p-norms illustrated.

	For any norm \|.\|, the dual norm is defined as \|y\|_* = max{ <x, y> for \|x\| <= 1 }.

	The figure shows the unit balls of the p-norm, for p = 1.5, 2, and 3.

	We compute the dual norm at a dual vector y (short black arrow), rotating
	uniformly around the origin over time.
	class Base:

	def say(self, val):
	print("base says", val)


	class A(Base):

	def say(self, val):
	print("say A")
	Example input and output.

	$ python conf.py seed=42 lr=.1

	project: ???
	seed: 42
	lr: 0.1
	epochs: ???
	p_drop: 0.5
	baseconf: null
	# author: vlad niculae <vlad@vene.ro>
	# license: mit

	import torch
	import numpy as np
	import matplotlib.pyplot as plt
	import matplotlib.colors as colors

	from entmax import sparsemax, entmax15
	from entmax.losses import sparsemax_loss, entmax15_loss
	# author: vn

	import numpy as np
	from scipy.optimize import root_scalar
	import torch
	import matplotlib.pyplot as plt


	def entropy(y, a, b):
	# Density estimation with energy-based models
	# Langevin sampling, contrastive divergence training.
	# Author: Vlad Niculae <vlad@vene.ro>
	# License: MIT

	import numpy as np
	import torch
	from sklearn import datasets
	import matplotlib.pyplot as plt
	# Geometric intepretation of the gradient of the mapping:
	# f : (0, inf) x Sphere(k-1) -> R^k
	# f(r, u) -> r*u
	# The catch: R can vary on (0, inf) but u may only vary on the
	# k-1--dimensional tangent plane!

	import numpy as np

	def main():