Disqus - Friends of shantanuo

Re: How to become a data scientist

Umar Yusuf — Wed, 03 May 2017 09:09:52 -0000

Nice article to say the least.

On a platform like dataquest, one should expect practical oriented guide not theoretical guide.

The choice is now for you to decide if you want practical or theory data science. If your choice is to go theory, then colleges and universities are there to study math, statistics, CS and related courses for 5years or more.

Re: Web Mapping with QGIS2Web — QGIS Tutorials and Tips

Umar Yusuf — Tue, 16 May 2017 19:12:17 -0000

You may have to create/update your "Photo" attribute field, so that it will hold the image in html tag (<img ...=""/>).

Re: The most popular Python scientific libraries

Umar Yusuf — Sun, 25 Jun 2017 05:58:55 -0000

Those in the Geosciences field will not be happy with the list above, here is a List of python Geoscience libraries:

~ GDAL –> Fundamental package for processing vector and raster data formats (many modules below depend on this). Used for raster processing.
~ Geopandas –> Working with geospatial data in Python made easier, combines the capabilities of pandas and shapely.
~ Shapely –> Python package for manipulation and analysis of planar geometric objects (based on widely deployed GEOS).
~ Fiona –> Reading and writing spatial data (alternative for geopandas).
~ Pyproj –> Performs cartographic transformations and geodetic computations (based on PROJ.4).
~ Pysal –> Library of spatial analysis functions written in Python.
~ Geopy –> Geocoding library: coordinates to address <-> address to coordinates.
~ GeoViews –> Interactive Maps for the web.
~ Cartopy –> Make drawing maps for data analysis and visualization as easy as possible.
~ Scipy.spatial –> Spatial algorithms and data structures.
~ Rtree –> Spatial indexing for Python for quick spatial lookups.
~ Rasterio –> Clean and fast and geospatial raster I/O for Python.
~ RSGISLib –> Remote Sensing and GIS Software Library for Python.

Source: https://umar-yusuf.blogspot...

Re: Launch your data science career in 5 steps (with Python)

Umar Yusuf — Mon, 24 Jul 2017 06:00:19 -0000

Thank you Kevin. This is very nice path to starting out a career in data science, keep it up.

Do you have useful links/material for someone more interested in Geo data science field? I mean data science that emphasis on Geographic datasets.

Re: Web scraping the President's lies in 16 lines of Python

Umar Yusuf — Fri, 28 Jul 2017 06:34:32 -0000

Nice tutorial. You kept it simple by not introducing re.

Re: wxPython Recipes Book Contest

Umar Yusuf — Thu, 01 Feb 2018 12:13:36 -0000

I came from a region of the world (Africa) where Python paperback copies are hardly seen in local book stores. If you send me a copy, it will be the very first am going to touch ever since I started learning python.

Re: File extension PGW - Software to open, view .pgw files

Umar Yusuf — Thu, 23 Aug 2018 11:08:18 -0000

When you geo-reference and save PNG image file of Panama for example, you will get the "world file" associated with it. I use QGIS software for this.

Re: 18 Hidden Things You Didn't Know Google Maps Could Do

Umar Yusuf — Mon, 05 Aug 2019 14:26:24 -0000

Awesome collection of my maps trick.

Re: Should you use "bracket notation" or "dot notation" with pandas?

Umar Yusuf — Sat, 14 Sep 2019 01:34:11 -0000

I prefer the 'bracket notation', just to keep consistent and avoid the issues you listed above...

test_df = pd.DataFrame(columns=['col_one', 'col_two', 'col_three', 'col_four'])
If 'dot notation' wont work with above, then why should I worry about just because it is easier to type!

Re: Learn a new pandas trick every day!

Umar Yusuf — Mon, 30 Dec 2019 03:56:24 -0000

Awesome collection of pandas tricks indeed. I wish to see something like this for all the major python data science libraries!

Re: Python Foundation for Spatial Analysis (Full Course Material)

Umar Yusuf — Tue, 26 Jan 2021 16:01:15 -0000

A nice introduction to GIS with Python.

Re: ghostery

Amrita — Tue, 16 Jun 2020 02:16:10 -0000

Can you expand on why you say there's "no term in the loss function that explicitly pulls different classes apart"? By maximizing the expected number of correctly classified points, we maximize the number of points whose most-probably-selected neighbor is of the correct class. Shouldn't that pull points within the same class closer together/push different classes apart in the transformed feature space?

Re: ghostery

Amrita — Tue, 16 Jun 2020 03:41:44 -0000

Thanks for your response! What's the magnitude on the y-axis? I'm trying to picture why the 2D embedding along that hyperplane/axis might still work well for KNN classification, even though it looks like the concentric-circles structure has been lost, almost like the starting point of the animation you have. It's definitely interesting to use the vanilla loss function as the "similar points" half of a contrastive loss and add a hinge loss for the "dissimilar points" component!