bnsh

#! /usr/bin/env python3
# vim: expandtab shiftwidth=4 tabstop=4

"""
    My application outputs a 2 dimensional quantity, but is batched.. I generate masks for the outputs that I want it to predict in the absence of having that data.

    The masking works. As in, I can do input[masks] = 0, and I can also read output[masks] to get the value of only the masked outputs and I can do
        loss = crit(output[masks], target[masks]) and get the loss for only the values that are in the mask.

    My issue is that I want to sum over the masks.. So, in the demo below, I have a batch size of 4 and

bnsh

# docker build -t binesh/pylint-bug .
# docker run --rm binesh/pylint-bug python3.8 -m pylint -r n /tmp/demo1.py
# docker run --rm binesh/pylint-bug python3.8 -m pylint -r n /tmp/demo2.py
FROM python:3.8

MAINTAINER Binesh Bannerjee <binesh_binesh@hotmail.com>

RUN /usr/local/bin/python3.8 -m pip install -U pylint==2.6.1

RUN echo "def func(blah):" > /tmp/demo1.py && \

bnsh

FROM ubuntu:20.04

LABEL maintainer="Binesh Bannerjee <binesh_binesh@hotmail.com>"

RUN \
	apt-get update -y \
	&& apt-get install -y curl gnupg software-properties-common \
	&& curl -fsSL https://apt.releases.hashicorp.com/gpg | apt-key add - \
	&& apt-add-repository -y "deb [arch=amd64] https://apt.releases.hashicorp.com $(lsb_release -cs) main" \
	&& apt-get update -y && apt-get install -y vault

bnsh

FROM pytorch/pytorch:latest
MAINTAINER Binesh Bannerjee <binesh_binesh@hotmail.com>
COPY test.py /tmp

bnsh

FROM python:3.8

MAINTAINER Binesh Bannerjee <binesh_binesh@hotmail.com>

RUN /usr/local/bin/python3.8 -m pip install -U pylint pandas

RUN echo "import pandas as pd" > /tmp/demo.py && \
    echo "pd.merge_asof()" >> /tmp/demo.py

RUN /usr/local/bin/python3.8 -m pip freeze

bnsh

#! /usr/bin/env python3
# vim: expandtab shiftwidth=4 tabstop=4

"""This program uses an xor network to test MyDataParallel"""

import argparse
from collections import OrderedDict
import torch
import torch.nn as nn
import torch.optim as optim

bnsh

/* vim: expandtab shiftwidth=4 tabstop=4
 */
/*
 * Given a min value and max value and a number of steps return an array of that many
 * points from min value to max value. min value _must_ be > 0.. Likely you want
 * to use 1 as min value. (the log of 0 is -Infinity, so you can't use 0...)
 */

function logmidpoints(min, max, steps) {
    var points = [];

bnsh

#include <torch/torch.h>
#include <iostream>
#include <vector>

at::Tensor d_sigmoid(at::Tensor z) {
	auto s = at::sigmoid(z);
	return (1-s) * s;
}

std::vector<at::Tensor> lltm_forward(

bnsh

#! /usr/bin/env python3

"""This is the MovieLens example from lightfm"""

import numpy as np
from scipy.sparse import csc_matrix
from lightfm.datasets import fetch_movielens
from lightfm import LightFM
from lightfm.evaluation import precision_at_k

bnsh

#! /usr/bin/env python3

"""Is this a bug in lightfm's precision_at_k function?

   If there are #entries < k for a particular user, it seems as if that user can _never_ have a precision
   > (#positive_interactions / k) for that user. So, the maximum precision is _not_ in fact 1.0, it's whatever the average
   of _all_ the (#positive_interactions / k) values are throughout the entire training set.

   I have a proposed solution at the end. I wonder if it works.
"""