Code Snippet Corner

The other day, I was grabbing a way-too-big DB query for local exploration.  It was literally over 2GB as a CSV - which is a pain for a number of reasons!  Not the least of which being that, while you're doing Exploratory Data Analysis, things take way too long - it doesn't take long for Cognitive Drift to break your rhythm!

Numerical columns can be taken down to size with the downcasting functions from a previous post.  But what about Object/String columns?  One of the best ways to reduce the size of a file like this is to recast

Sometimes you wind up with duplicate column names in your Pandas DataFrame. This isn't necessarily a huge deal if we're just messing with a smallish file in Jupyter.  But, if we wanna do something like load it into a database, that'll be a problem.  It can also interfere with our other cleaning functions - I ran into this the other day when reducing the size of a giant data file by downcasting it (as per this previous post).  The cleaning functions required a 1D input (so, a Series or List) - but calling the name of a duplicate column gave

Notebook here.  Helper functions here.

One of the best features of Random Forests is that it has built-in Feature Selection.  Explicability is one of the things we often lose when we go from traditional statistics to Machine Learning, but Random Forests lets us actually get some insight into our dataset instead of just having to treat our model as a black box.

One problem, though - it doesn't work that well for categorical features.  Since you'll generally have to One-Hot Encode a categorical feature (for instance, turn something with 7 categories into 7 variables that are a "True/False"), you'll

Part 1 (n_estimators) here
Part 2 (max_depth) here
Notebook here

Another parameter, another set of quirks!

min_samples_leaf is sort of similar to max_depth.  It helps us avoid overfitting.  It's also non-obvious what you should use as your upper and lower limits to search between.  Let's do what we did last week - build a forest with no parameters, see what it does, and use the upper and lower limits!

import pandas as pd

from sklearn.metrics import precision_recall_curve
from sklearn.metrics import auc
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_

Continued from here.

Notebook for this post is here.

Binary search code itself is here.

max_depth is an interesting parameter.  While n_estimators has a tradeoff between speed & score, max_depth has the possibility of improving both.  By limiting the depth of your trees, you can reduce overfitting.

Unfortunately, deciding on upper & lower bounds is less than straightforward.  It'll depend on your dataset.  Luckily, I found a post on StackOverflow that had a link to a blog post that had a promising methodology.  

First, we build a tree with default arguments and fit it to our data.

Ah, hyperparameter tuning.  Time & compute-intensive.  Frequently containing weird non-linearities in how changing a parameter changes the score and/or the time it takes to train the model.

RandomizedSearchCV goes noticeably faster than a full GridSearchCV but it still takes a while - which can be rough, because in my experience you do still need to be iterative with it and experiment with different distributions.  Plus, then you've got hyper-hyperparameters to tune - how many iterations SHOULD you run it for, anyway?

I've been experimenting with using the trusty old Binary Search to tune hyperparameters.  I'm finding it has two

What if, like during my data import task a few months back, the dates & times are in separate columns?  This gives us a few new issues.  Let's import that Excel file!

import pandas as pd
import xlrd
import datetime

df = pd.read_excel("hasDatesAndTimes.xlsx", sheet_name="Sheet1")

book = xlrd.open_workbook("hasDatesAndTimes.xlsx")
datemode = book.datemode

And let's see that time variable!

df["Time"]

df["Time"].map(lambda x: xlrd.xldate_

Different file formats are different!  For all kinds of reasons!

A few months back, I had to import some Excel files into a database. In this process I learned so much about the delightfully unique way Excel stores dates & times!  

The basic datetime will be a decimal number, like 43324.909907407404.  The number before the decimal is the day, the number afterwards is the time.  So far, so good - this is pretty common for computers.  The date is often the number of days past a certain date, and the time is the number of seconds.  

So, let's load

Ah, laziness.  You love it, I love it, everyone agrees it's just better.

Flesh-and-blood are famously lazy.  Pandas the package, however, uses Eager Evaluation.  What's Eager Evaluation, you ask?  Is Pandas really judgey, hanging out on the street corner and being fierce to the style choices of people walking by?  Well, yes, but that's not the most relevant sense in which I mean it here.  

Eager evaluation means that once you call pd.read_csv(), Pandas immediately jumps to read the whole CSV into memory.

"Wait!" I hear you ask.  "Isn't that what we want?  Why would I call the

Last few Code Snippet Corners were about using Pandas as an easy way to handle input and output between files & databases.  Let's shift gears a little bit!  Among other reasons, because earlier today I discovered a package that exclusively does that, which means I can stop importing the massive Pandas package when all I really wanted to do with it was take advantage of its I/O modules.  Check it out!

So, rather than the entrances & exits, let's focus on all the crazy ways you can reshape data with Pandas!

Our Data

For our demonstration, I'll use a

This isn't really a tutorial on cron in general; Better people at Linux have written way better ones than I could write.  Here's one: http://mediatemple.net/blog/news/complete-beginners-guide-cron-part-1/  This is more of a code journaling exercise for a problem that I didn't find a neat-and-tidy answer to online when I was looking for it, and that I presume at least one person will encounter at some point between now and the heat death of the universe.

Let's say you've got two different Conda envs:  production and development.  Let's say that, in addition to having different packages installed, they