Nikolay Lysenko Nikolay-Lysenko

The Problem

Suppose that Jupyter notebooks are hosted at a remote server and you access them from your local machine via web interface. By default, if connection is closed (e.g., you close a browser tab with remotely hosted notebook), corresponding to this connection running kernels die. Hence, time-consuming processes can not be finished without maintaining their connections open. Sometimes it is unfeasible — for instance, assume that you must shut down your local machine in the evening and you can turn it on only in the next morning.

The Solution

Initial Once-Only Preparations

Open local .bashrc file in edit mode:

	"""
	Implementation of non-equi inner join in `pandas`.
	Non-equi join is a join with conditions other than exact match.
	For example, values from a column of a left `DataFrame` must be higher
	than values from a column of a right `DataFrame` and lower than values
	from another column of the right `DataFrame`.

	@author: Nikolay Lysenko
	"""

	"""
	This script demonstrates why train set and
	test set must not be shuffled if a dataset has
	temporal structure. In case of temporal
	structure, shuffling leads to leakage of some
	information about a test set to a training set
	and this results in too optimistic scores.

	The point of the considered here example is that
	a particular flaw of tree-based ensembles

	from time import sleep
	from io import StringIO

	import psycopg2


	def upsert_df_into_postgres(df, target_table, primary_keys, conn_string,
	n_trials=5, quoting=None, null_repr=None):
	"""
	Uploads data from `df` to `target_table`

	import numpy as np


	class TwoLayerPerceptron(object):
	"""
	This is a simple neural network
	with exactly one hidden layer
	and 0.5 * MSE (Mean Squared Error)
	as loss.

	import numpy as np


	def xgb_quantile_eval(preds, dmatrix, quantile=0.2):
	"""
	Customized evaluational metric that equals
	to quantile regression loss (also known as
	pinball loss).

	Quantile regression is regression that