01-numpy.md 28.3 KB
Newer Older
1
---
2
3
4
5
title: Analyzing Patient Data
teaching: 30
exercises: 0
questions:
Greg Wilson's avatar
Greg Wilson committed
6
- "How can I process tabular data files in Python?"
7
8
9
10
11
12
13
14
15
objectives:
- "Explain what a library is, and what libraries are used for."
- "Import a Python library and use the things it contains."
- "Read tabular data from a file into a program."
- "Assign values to variables."
- "Select individual values and subsections from data."
- "Perform operations on arrays of data."
- "Display simple graphs."
keypoints:
Greg Wilson's avatar
Greg Wilson committed
16
17
18
19
20
21
22
23
24
25
26
27
28
29
- "Import a library into a program using `import libraryname`."
- "Use the `numpy` library to work with arrays in Python."
- "Use `variable = value` to assign a value to a variable in order to record it in memory."
- "Variables are created on demand whenever a value is assigned to them."
- "Use `print(something)` to display the value of `something`."
- "The expression `array.shape` gives the shape of an array."
- "Use `array[x, y]` to select a single element from an array."
- "Array indices start at 0, not 1."
- "Use `low:high` to specify a slice that includes the indices from `low` to `high-1`."
- "All the indexing and slicing that works on arrays also works on strings."
- "Use `# some kind of explanation` to add comments to programs."
- "Use `numpy.mean(array)`, `numpy.max(array)`, and `numpy.min(array)` to calculate simple statistics."
- "Use `numpy.mean(array, axis=0)` or `numpy.mean(array, axis=1)` to calculate statistics across the specified axis."
- "Use the `pyplot` library from `matplotlib` for creating simple visualizations."
30
31
---

32
Words are useful,
33
but what's more useful are the sentences and stories we build with them.
34
Similarly,
jstapleton's avatar
jstapleton committed
35
while a lot of powerful, general tools are built into languages like Python,
36
specialized tools built up from these basic units live in [libraries](reference.html#library)
jstapleton's avatar
jstapleton committed
37
that can be called upon when needed.
38
39

In order to load our inflammation data,
Trevor Bekolay's avatar
Trevor Bekolay committed
40
41
we need to access ([import](reference.html#import) in Python terminology)
a library called [NumPy](http://docs.scipy.org/doc/numpy/ "NumPy Documentation").
Greg Wilson's avatar
Greg Wilson committed
42
In general you should use this library if you want to do fancy things with numbers,
43
especially if you have matrices or arrays.
Trevor Bekolay's avatar
Trevor Bekolay committed
44
We can import NumPy using:
45

46
~~~
Greg Wilson's avatar
Greg Wilson committed
47
48
import numpy
~~~
49
{: .python}
50

51
52
53
Importing a library is like getting a piece of lab equipment out of a storage locker and setting it up on the bench.
Libraries provide additional functionality to the basic Python package,
much like a new piece of equipment adds functionality to a lab space.
54
Once you've imported the library,
55
we can ask the library to read our data file for us:
56

57
~~~
Peter Cock's avatar
Peter Cock committed
58
numpy.loadtxt(fname='inflammation-01.csv', delimiter=',')
Greg Wilson's avatar
Greg Wilson committed
59
~~~
60
61
62
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
63
array([[ 0.,  0.,  1., ...,  3.,  0.,  0.],
64
65
       [ 0.,  1.,  2., ...,  1.,  0.,  1.],
       [ 0.,  1.,  1., ...,  2.,  1.,  1.],
66
       ...,
67
68
       [ 0.,  1.,  1., ...,  1.,  1.,  1.],
       [ 0.,  0.,  0., ...,  0.,  2.,  0.],
Greg Wilson's avatar
Greg Wilson committed
69
70
       [ 0.,  0.,  1., ...,  1.,  1.,  0.]])
~~~
71
{: .output}
72

Greg Wilson's avatar
Greg Wilson committed
73
The expression `numpy.loadtxt(...)` is a [function call](reference.html#function-call)
74
that asks Python to run the [function](reference.html#function) `loadtxt` which belongs to the `numpy` library.
Greg Wilson's avatar
Greg Wilson committed
75
This [dotted notation](reference.html#dotted-notation) is used everywhere in Python
76
to refer to the parts of things as `thing.component`.
77

Greg Wilson's avatar
Greg Wilson committed
78
`numpy.loadtxt` has two [parameters](reference.html#parameter):
79
the name of the file we want to read,
Greg Wilson's avatar
Greg Wilson committed
80
81
and the [delimiter](reference.html#delimiter) that separates values on a line.
These both need to be character strings (or [strings](reference.html#string) for short),
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
so we put them in quotes.

When we are finished typing and press Shift+Enter,
the notebook runs our command.
Since we haven't told it to do anything else with the function's output,
the notebook displays it.
In this case,
that output is the data we just loaded.
By default,
only a few rows and columns are shown
(with `...` to omit elements when displaying big arrays).
To save space,
Python displays numbers as `1.` instead of `1.0`
when there's nothing interesting after the decimal point.

Our call to `numpy.loadtxt` read our file,
but didn't save the data in memory.
To do that,
Greg Wilson's avatar
Greg Wilson committed
100
we need to [assign](reference.html#assignment) the array to a [variable](reference.html#variable).
101
A variable is just a name for a value,
Greg Wilson's avatar
Greg Wilson committed
102
such as `x`, `current_temperature`, or `subject_id`.
103
Python's variables must begin with a letter and are [case sensitive](reference.html#case-sensitive).
Kyler Brown's avatar
Kyler Brown committed
104
We can create a new variable by assigning a value to it using `=`.
Greg Wilson's avatar
Greg Wilson committed
105
106
107
108
As an illustration,
let's step back and instead of considering a table of data,
consider the simplest "collection" of data,
a single value.
109
The line below assigns the value `55` to a variable `weight_kg`:
Greg Wilson's avatar
Greg Wilson committed
110

111
~~~
Greg Wilson's avatar
Greg Wilson committed
112
113
weight_kg = 55
~~~
114
{: .python}
115

116
Once a variable has a value, we can print it to the screen:
117

118
~~~
119
print(weight_kg)
Greg Wilson's avatar
Greg Wilson committed
120
~~~
121
122
123
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
124
125
55
~~~
126
{: .output}
127

128
and do arithmetic with it:
129

130
~~~
131
print('weight in pounds:', 2.2 * weight_kg)
Greg Wilson's avatar
Greg Wilson committed
132
~~~
133
134
135
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
136
137
weight in pounds: 121.0
~~~
138
{: .output}
139

jstapleton's avatar
jstapleton committed
140
141
142
As the example above shows,
we can print several things at once by separating them with commas.

143
We can also change a variable's value by assigning it a new one:
144

145
~~~
Greg Wilson's avatar
Greg Wilson committed
146
weight_kg = 57.5
147
print('weight in kilograms is now:', weight_kg)
Greg Wilson's avatar
Greg Wilson committed
148
~~~
149
150
151
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
152
153
weight in kilograms is now: 57.5
~~~
154
{: .output}
155

156
157
If we imagine the variable as a sticky note with a name written on it,
assignment is like putting the sticky note on a particular value:
158

159
![Variables as Sticky Notes]({{ site.github.url }}/fig/python-sticky-note-variables-01.svg)
160

161
162
163
This means that assigning a value to one variable does *not* change the values of other variables.
For example,
let's store the subject's weight in pounds in a variable:
164

165
~~~
Greg Wilson's avatar
Greg Wilson committed
166
weight_lb = 2.2 * weight_kg
167
print('weight in kilograms:', weight_kg, 'and in pounds:', weight_lb)
Greg Wilson's avatar
Greg Wilson committed
168
~~~
169
170
171
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
172
173
weight in kilograms: 57.5 and in pounds: 126.5
~~~
174
{: .output}
175

176
![Creating Another Variable]({{ site.github.url }}/fig/python-sticky-note-variables-02.svg)
177

178
and then change `weight_kg`:
179

180
~~~
Greg Wilson's avatar
Greg Wilson committed
181
weight_kg = 100.0
182
print('weight in kilograms is now:', weight_kg, 'and weight in pounds is still:', weight_lb)
Greg Wilson's avatar
Greg Wilson committed
183
~~~
184
185
186
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
187
188
weight in kilograms is now: 100.0 and weight in pounds is still: 126.5
~~~
189
{: .output}
190

191
![Updating a Variable]({{ site.github.url }}/fig/python-sticky-note-variables-03.svg)
192

193
194
195
196
Since `weight_lb` doesn't "remember" where its value came from,
it isn't automatically updated when `weight_kg` changes.
This is different from the way spreadsheets work.

197
> ## Who's Who in Memory
Benjamin Laken's avatar
Benjamin Laken committed
198
>
199
200
201
> You can use the `%whos` command at any time to see what
> variables you have created and what modules you have loaded into the computer's memory.
> As this is an IPython command, it will only work if you are in an IPython terminal or the Jupyter Notebook.
Benjamin Laken's avatar
Benjamin Laken committed
202
>
203
> ~~~
204
205
> %whos
> ~~~
206
207
208
> {: .python}
>
> ~~~
209
210
211
212
213
214
> Variable    Type       Data/Info
> --------------------------------
> numpy       module     <module 'numpy' from '/Us<...>kages/numpy/__init__.py'>
> weight_kg   float      100.0
> weight_lb   float      126.5
> ~~~
215
216
> {: .output}
{: .callout}
Benjamin Laken's avatar
Benjamin Laken committed
217

Johnny Lin's avatar
Johnny Lin committed
218
219
Just as we can assign a single value to a variable, we can also assign an array of values
to a variable using the same syntax.  Let's re-run `numpy.loadtxt` and save its result:
220

221
~~~
Greg Wilson's avatar
Greg Wilson committed
222
223
data = numpy.loadtxt(fname='inflammation-01.csv', delimiter=',')
~~~
224
{: .python}
225

226
227
228
This statement doesn't produce any output because assignment doesn't display anything.
If we want to check that our data has been loaded,
we can print the variable's value:
229

230
~~~
231
print(data)
Greg Wilson's avatar
Greg Wilson committed
232
~~~
233
234
235
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
236
[[ 0.  0.  1. ...,  3.  0.  0.]
237
238
 [ 0.  1.  2. ...,  1.  0.  1.]
 [ 0.  1.  1. ...,  2.  1.  1.]
239
 ...,
240
241
242
 [ 0.  1.  1. ...,  1.  1.  1.]
 [ 0.  0.  0. ...,  0.  2.  0.]
 [ 0.  0.  1. ...,  1.  1.  0.]]
Greg Wilson's avatar
Greg Wilson committed
243
~~~
244
{: .output}
245

246
247
248
Now that our data is in memory,
we can start doing things with it.
First,
Greg Wilson's avatar
Greg Wilson committed
249
let's ask what [type](reference.html#type) of thing `data` refers to:
250

251
~~~
252
print(type(data))
Greg Wilson's avatar
Greg Wilson committed
253
~~~
254
255
256
{: .python}

~~~
257
<class 'numpy.ndarray'>
Greg Wilson's avatar
Greg Wilson committed
258
~~~
259
{: .output}
260

261
262
263
264
265
266
The output tells us that `data` currently refers to
an N-dimensional array created by the NumPy library.
These data correspond to arthritis patients' inflammation.
The rows are the individual patients and the columns
are their daily inflammation measurements.

267
> ## Data Type
268
269
270
271
>
> A Numpy array contains one or more elements
> of the same type. `type` will only tell you that
> a variable is a NumPy array.
Valentina Staneva's avatar
Valentina Staneva committed
272
> We can also find out the type
273
274
> of the data contained in the NumPy array.
>
275
> ~~~
276
277
> print(data.dtype)
> ~~~
278
279
280
> {: .python}
>
> ~~~
281
282
> dtype('float64')
> ~~~
283
> {: .output}
284
285
286
>
> This tells us that the NumPy array's elements are
> [floating-point numbers](reference.html#floating-point number).
287
{: .callout}
288

jstapleton's avatar
jstapleton committed
289
We can see what the array's [shape](reference.html#shape) is like this:
290

291
~~~
292
print(data.shape)
Greg Wilson's avatar
Greg Wilson committed
293
~~~
294
295
296
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
297
298
(60, 40)
~~~
299
{: .output}
300

301
302
This tells us that `data` has 60 rows and 40 columns. When we created the
variable `data` to store our arthritis data, we didn't just create the array, we also
Trevor Bekolay's avatar
Trevor Bekolay committed
303
created information about the array, called [members](reference.html#member) or
304
305
attributes. This extra information describes `data` in
the same way an adjective describes a noun.
Bartosz T's avatar
Bartosz T committed
306
`data.shape` is an attribute  of `data` which describes the dimensions of `data`.
307
We use the same dotted notation for the attributes of variables
308
309
that we use for the functions in libraries
because they have the same part-and-whole relationship.
310

311
If we want to get a single number from the array,
Greg Wilson's avatar
Greg Wilson committed
312
we must provide an [index](reference.html#index) in square brackets,
313
just as we do in math:
314

315
~~~
316
print('first value in data:', data[0, 0])
Greg Wilson's avatar
Greg Wilson committed
317
~~~
318
319
320
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
321
322
first value in data: 0.0
~~~
323
{: .output}
324

325
~~~
326
print('middle value in data:', data[30, 20])
Greg Wilson's avatar
Greg Wilson committed
327
~~~
328
329
330
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
331
332
middle value in data: 13.0
~~~
333
{: .output}
334

335
336
337
338
339
The expression `data[30, 20]` may not surprise you,
but `data[0, 0]` might.
Programming languages like Fortran and MATLAB start counting at 1,
because that's what human beings have done for thousands of years.
Languages in the C family (including C++, Java, Perl, and Python) count from 0
340
341
342
because that's more convenient when indices are computed rather than constant
(see [Mike Hoye's blog post](http://exple.tive.org/blarg/2013/10/22/citation-needed/)
for historical details).
343
As a result,
Greg Wilson's avatar
Greg Wilson committed
344
if we have an M×N array in Python,
345
346
347
348
349
350
its indices go from 0 to M-1 on the first axis
and 0 to N-1 on the second.
It takes a bit of getting used to,
but one way to remember the rule is that
the index is how many steps we have to take from the start to get the item we want.

351
> ## In the Corner
352
353
354
355
356
357
>
> What may also surprise you is that when Python displays an array,
> it shows the element with index `[0, 0]` in the upper left corner
> rather than the lower left.
> This is consistent with the way mathematicians draw matrices,
> but different from the Cartesian coordinates.
358
> The indices are (row, column) instead of (column, row) for the same reason,
359
> which can be confusing when plotting data.
360
{: .callout}
361
362
363
364
365

An index like `[30, 20]` selects a single element of an array,
but we can select whole sections as well.
For example,
we can select the first ten days (columns) of values
366
for the first four patients (rows) like this:
367

368
~~~
369
print(data[0:4, 0:10])
Greg Wilson's avatar
Greg Wilson committed
370
~~~
371
372
373
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
374
[[ 0.  0.  1.  3.  1.  2.  4.  7.  8.  3.]
375
376
377
 [ 0.  1.  2.  1.  2.  1.  3.  2.  2.  6.]
 [ 0.  1.  1.  3.  3.  2.  6.  2.  5.  9.]
 [ 0.  0.  2.  0.  4.  2.  2.  1.  6.  7.]]
Greg Wilson's avatar
Greg Wilson committed
378
~~~
379
{: .output}
380

Greg Wilson's avatar
Greg Wilson committed
381
The [slice](reference.html#slice) `0:4` means,
382
383
384
385
386
387
"Start at index 0 and go up to, but not including, index 4."
Again,
the up-to-but-not-including takes a bit of getting used to,
but the rule is that the difference between the upper and lower bounds is the number of values in the slice.

We don't have to start slices at 0:
388

389
~~~
390
print(data[5:10, 0:10])
Greg Wilson's avatar
Greg Wilson committed
391
~~~
392
393
394
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
395
[[ 0.  0.  1.  2.  2.  4.  2.  1.  6.  4.]
396
397
398
399
 [ 0.  0.  2.  2.  4.  2.  2.  5.  5.  8.]
 [ 0.  0.  1.  2.  3.  1.  2.  3.  5.  3.]
 [ 0.  0.  0.  3.  1.  5.  6.  5.  5.  8.]
 [ 0.  1.  1.  2.  1.  3.  5.  3.  5.  8.]]
Greg Wilson's avatar
Greg Wilson committed
400
~~~
401
{: .output}
402

403
404
405
406
407
408
409
410
We also don't have to include the upper and lower bound on the slice.
If we don't include the lower bound,
Python uses 0 by default;
if we don't include the upper,
the slice runs to the end of the axis,
and if we don't include either
(i.e., if we just use ':' on its own),
the slice includes everything:
411

412
~~~
Greg Wilson's avatar
Greg Wilson committed
413
small = data[:3, 36:]
414
415
print('small is:')
print(small)
Greg Wilson's avatar
Greg Wilson committed
416
~~~
417
418
419
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
420
small is:
421
422
423
[[ 2.  3.  0.  0.]
 [ 1.  1.  0.  1.]
 [ 2.  2.  1.  1.]]
Greg Wilson's avatar
Greg Wilson committed
424
~~~
425
{: .output}
426

427
Arrays also know how to perform common mathematical operations on their values.
Greg Wilson's avatar
Greg Wilson committed
428
429
430
431
432
The simplest operations with data are arithmetic:
add, subtract, multiply, and divide.
 When you do such operations on arrays,
the operation is done on each individual element of the array.
Thus:
433

434
~~~
Greg Wilson's avatar
Greg Wilson committed
435
436
doubledata = data * 2.0
~~~
437
{: .python}
438

Greg Wilson's avatar
Greg Wilson committed
439
440
will create a new array `doubledata`
whose elements have the value of two times the value of the corresponding elements in `data`:
441

442
~~~
443
444
445
446
print('original:')
print(data[:3, 36:])
print('doubledata:')
print(doubledata[:3, 36:])
Greg Wilson's avatar
Greg Wilson committed
447
~~~
448
449
450
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
451
original:
452
453
454
455
456
457
458
[[ 2.  3.  0.  0.]
 [ 1.  1.  0.  1.]
 [ 2.  2.  1.  1.]]
doubledata:
[[ 4.  6.  0.  0.]
 [ 2.  2.  0.  2.]
 [ 4.  4.  2.  2.]]
Greg Wilson's avatar
Greg Wilson committed
459
~~~
460
{: .output}
461

Greg Wilson's avatar
Greg Wilson committed
462
463
If,
instead of taking an array and doing arithmetic with a single value (as above)
464
you did the arithmetic operation with another array of the same shape,
Greg Wilson's avatar
Greg Wilson committed
465
466
the operation will be done on corresponding elements of the two arrays.
Thus:
467

468
~~~
Greg Wilson's avatar
Greg Wilson committed
469
470
tripledata = doubledata + data
~~~
471
{: .python}
472

473
474
475
will give you an array where `tripledata[0,0]` will equal `doubledata[0,0]` plus `data[0,0]`,
and so on for all other elements of the arrays.

476
~~~
477
478
print('tripledata:')
print(tripledata[:3, 36:])
Greg Wilson's avatar
Greg Wilson committed
479
~~~
480
481
482
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
483
tripledata:
484
485
486
[[ 6.  9.  0.  0.]
 [ 3.  3.  0.  3.]
 [ 6.  6.  3.  3.]]
Greg Wilson's avatar
Greg Wilson committed
487
~~~
488
{: .output}
489

490
Often, we want to do more than add, subtract, multiply, and divide values of data.
491
NumPy knows how to do more complex operations on arrays.
492
493
If we want to find the average inflammation for all patients on all days,
for example,
494
we can ask NumPy to compute `data`'s mean value:
495

496
~~~
497
print(numpy.mean(data))
Greg Wilson's avatar
Greg Wilson committed
498
~~~
499
500
501
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
502
503
6.14875
~~~
504
{: .output}
505

506
507
508
509
510
`mean` is a [function](reference.html#function) that takes
an array as an [argument](reference.html#argument).
If variables are nouns, functions are verbs:
they do things with variables.

511
> ## Not All Functions Have Input
512
513
514
>
> Generally, a function uses inputs to produce outputs.
> However, some functions produce outputs without
515
516
> needing any input. For example, checking the current time
> doesn't require any input.
517
>
518
> ~~~
519
520
> import time
> print(time.ctime())
521
> ~~~
522
523
524
> {: .python}
>
> ~~~
525
> 'Sat Mar 26 13:07:33 2016'
526
> ~~~
527
> {: .output}
528
529
530
531
>
> For functions that don't take in any arguments,
> we still need parentheses (`()`)
> to tell Python to go and do something for us.
532
{: .callout}
533
534
535

NumPy has lots of useful functions that take an array as input.
Let's use three of those functions to get some descriptive values about the dataset.
536
537
We'll also use multiple assignment,
a convenient Python feature that will enable us to do this all in one line.
538

539
~~~
540
maxval, minval, stdval = numpy.max(data), numpy.min(data), numpy.std(data)
541

Alistair Walsh's avatar
Alistair Walsh committed
542
543
544
print('maximum inflammation:', maxval)
print('minimum inflammation:', minval)
print('standard deviation:', stdval)
Greg Wilson's avatar
Greg Wilson committed
545
~~~
546
547
548
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
549
maximum inflammation: 20.0
550
551
minimum inflammation: 0.0
standard deviation: 4.61383319712
Greg Wilson's avatar
Greg Wilson committed
552
~~~
553
{: .output}
554

555
> ## Mystery Functions in IPython
556
>
557
558
559
560
561
562
563
> How did we know what functions NumPy has and how to use them?
> If you are working in the IPython/Jupyter Notebook there is an easy way to find out.
> If you type the name of something with a full-stop then you can use tab completion
> (e.g. type `numpy.` and then press tab)
> to see a list of all functions and attributes that you can use. After selecting one you
> can also add a question mark (e.g. `numpy.cumprod?`) and IPython will return an
> explanation of the method! This is the same as doing `help(numpy.cumprod)`.
564
{: .callout}
565
566

When analyzing data, though,
567
568
569
we often want to look at partial statistics,
such as the maximum value per patient
or the average value per day.
570
One way to do this is to create a new temporary array of the data we want,
571
then ask it to do the calculation:
572

573
~~~
Greg Wilson's avatar
Greg Wilson committed
574
patient_0 = data[0, :] # 0 on the first axis, everything on the second
575
print('maximum inflammation for patient 0:', patient_0.max())
Greg Wilson's avatar
Greg Wilson committed
576
~~~
577
578
579
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
580
581
maximum inflammation for patient 0: 18.0
~~~
582
{: .output}
583

584
585
586
Everything in a line of code following the '#' symbol is a
[comment](reference.html#comment) that is ignored by the computer.
Comments allow programmers to leave explanatory notes for other
jstapleton's avatar
jstapleton committed
587
588
programmers or their future selves.

589
We don't actually need to store the row in a variable of its own.
590
Instead, we can combine the selection and the function call:
591

592
~~~
593
print('maximum inflammation for patient 2:', numpy.max(data[2, :]))
Greg Wilson's avatar
Greg Wilson committed
594
~~~
595
596
597
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
598
599
maximum inflammation for patient 2: 19.0
~~~
600
{: .output}
601

602
What if we need the maximum inflammation for each patient over all days (as in the
603
604
605
next diagram on the left), or the average for each day (as in the
diagram on the right)? As the diagram below shows, we want to perform the
operation across an axis:
606

607
![Operations Across Axes]({{ site.github.url }}/fig/python-operations-across-axes.png)
608

609
To support this,
610
most array functions allow us to specify the axis we want to work on.
611
If we ask for the average across axis 0 (rows in our 2D example),
612
we get:
613

614
~~~
615
print(numpy.mean(data, axis=0))
Greg Wilson's avatar
Greg Wilson committed
616
~~~
617
618
619
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
620
[  0.           0.45         1.11666667   1.75         2.43333333   3.15
621
622
623
624
625
626
627
   3.8          3.88333333   5.23333333   5.51666667   5.95         5.9
   8.35         7.73333333   8.36666667   9.5          9.58333333
  10.63333333  11.56666667  12.35        13.25        11.96666667
  11.03333333  10.16666667  10.           8.66666667   9.15         7.25
   7.33333333   6.58333333   6.06666667   5.95         5.11666667   3.6
   3.3          3.56666667   2.48333333   1.5          1.13333333
   0.56666667]
Greg Wilson's avatar
Greg Wilson committed
628
~~~
629
{: .output}
630

631
632
As a quick check,
we can ask this array what its shape is:
633

634
~~~
635
print(numpy.mean(data, axis=0).shape)
Greg Wilson's avatar
Greg Wilson committed
636
~~~
637
638
639
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
640
641
(40,)
~~~
642
{: .output}
643

Greg Wilson's avatar
Greg Wilson committed
644
The expression `(40,)` tells us we have an N×1 vector,
645
so this is the average inflammation per day for all patients.
646
If we average across axis 1 (columns in our 2D example), we get:
647

648
~~~
649
print(numpy.mean(data, axis=1))
Greg Wilson's avatar
Greg Wilson committed
650
~~~
651
652
653
{: .python}

~~~
Greg Wilson's avatar
Greg Wilson committed
654
[ 5.45   5.425  6.1    5.9    5.55   6.225  5.975  6.65   6.625  6.525
655
656
657
658
659
  6.775  5.8    6.225  5.75   5.225  6.3    6.55   5.7    5.85   6.55
  5.775  5.825  6.175  6.1    5.8    6.425  6.05   6.025  6.175  6.55
  6.175  6.35   6.725  6.125  7.075  5.725  5.925  6.15   6.075  5.75
  5.975  5.725  6.3    5.9    6.75   5.925  7.225  6.15   5.95   6.275  5.7
  6.1    6.825  5.975  6.725  5.7    6.25   6.4    7.05   5.9  ]
Greg Wilson's avatar
Greg Wilson committed
660
~~~
661
{: .output}
662

663
664
665
666
667
which is the average inflammation per patient across all days.

The mathematician Richard Hamming once said,
"The purpose of computing is insight, not numbers,"
and the best way to develop insight is often to visualize data.
Valentina Staneva's avatar
Valentina Staneva committed
668
Visualization deserves an entire lecture (of course) of its own,
669
but we can explore a few features of Python's `matplotlib` library here.
Greg Wilson's avatar
Greg Wilson committed
670
671
While there is no "official" plotting library,
this package is the de facto standard.
672
673
674
First,
we will import the `pyplot` module from `matplotlib`
and use two of its functions to create and display a heat map of our data:
675

676
~~~
677
678
import matplotlib.pyplot
image  = matplotlib.pyplot.imshow(data)
679
matplotlib.pyplot.show()
Greg Wilson's avatar
Greg Wilson committed
680
~~~
681
{: .python}
682

683
![Heatmap of the Data]({{ site.github.url }}/fig/01-numpy_71_0.png)
684

685
686
687
Blue regions in this heat map are low values, while red shows high values.
As we can see,
inflammation rises and falls over a 40-day period.
688

689
> ## Some IPython Magic
690
691
692
693
>
> If you're using an IPython / Jupyter notebook,
> you'll need to execute the following command
> in order for your matplotlib images to appear
Damien Irving's avatar
Damien Irving committed
694
> in the notebook when `show()` is called:
Damien Irving's avatar
Damien Irving committed
695
>
696
> ~~~
Damien Irving's avatar
Damien Irving committed
697
698
> % matplotlib inline
> ~~~
699
> {: .python}
700
>
701
702
> The `%` indicates an IPython magic function -
> a function that is only valid within the notebook environment.
703
> Note that you only have to execute this function once per notebook.
704
{: .callout}
705

706
Let's take a look at the average inflammation over time:
707

708
~~~
709
ave_inflammation = numpy.mean(data, axis=0)
710
ave_plot = matplotlib.pyplot.plot(ave_inflammation)
711
matplotlib.pyplot.show()
Greg Wilson's avatar
Greg Wilson committed
712
~~~
713
{: .python}
714

715
![Average Inflammation Over Time]({{ site.github.url }}/fig/01-numpy_73_0.png)
716

717
718
Here,
we have put the average per day across all patients in the variable `ave_inflammation`,
719
then asked `matplotlib.pyplot` to create and display a line graph of those values.
720
721
722
723
724
The result is roughly a linear rise and fall,
which is suspicious:
based on other studies,
we expect a sharper rise and slower fall.
Let's have a look at two other statistics:
725

726
~~~
727
max_plot = matplotlib.pyplot.plot(numpy.max(data, axis=0))
728
matplotlib.pyplot.show()
Greg Wilson's avatar
Greg Wilson committed
729
~~~
730
{: .python}
731

732
![Maximum Value Along The First Axis]({{ site.github.url }}/fig/01-numpy_75_1.png)
Greg Wilson's avatar
Greg Wilson committed
733

734
~~~
735
min_plot = matplotlib.pyplot.plot(numpy.min(data, axis=0))
736
matplotlib.pyplot.show()
Greg Wilson's avatar
Greg Wilson committed
737
~~~
738
{: .python}
739

740
![Minimum Value Along The First Axis]({{ site.github.url }}/fig/01-numpy_75_3.png)
741

742
743
744
745
746
The maximum value rises and falls perfectly smoothly,
while the minimum seems to be a step function.
Neither result seems particularly likely,
so either there's a mistake in our calculations
or something is wrong with our data.
747
This insight would have been difficult to reach by
jstapleton's avatar
jstapleton committed
748
examining the data without visualization tools.
749

750
You can group similar plots in a single figure using subplots.
751
This script below uses a number of new commands. The function `matplotlib.pyplot.figure()`
752
creates a space into which we will place all of our plots. The parameter `figsize`
753
tells Python how big to make this space. Each subplot is placed into the figure using
754
its `add_subplot` [method](reference.html#method). The `add_subplot` method takes 3 parameters. The first denotes
755
how many total rows of subplots there are, the second parameter refers to the
valiseverywhere's avatar
valiseverywhere committed
756
total number of subplot columns, and the final parameter denotes which subplot
757
758
759
your variable is referencing (left-to-right, top-to-bottom). Each subplot is stored in a
different variable (`axes1`, `axes2`, `axes3`). Once a subplot is created, the axes can
be titled using the `set_xlabel()` command (or `set_ylabel()`).
760
Here are our three plots side by side:
761

762
~~~
763
764
import numpy
import matplotlib.pyplot
765

766
data = numpy.loadtxt(fname='inflammation-01.csv', delimiter=',')
767

768
fig = matplotlib.pyplot.figure(figsize=(10.0, 3.0))
769

770
771
772
axes1 = fig.add_subplot(1, 3, 1)
axes2 = fig.add_subplot(1, 3, 2)
axes3 = fig.add_subplot(1, 3, 3)
773

774
axes1.set_ylabel('average')
775
axes1.plot(numpy.mean(data, axis=0))
776

777
axes2.set_ylabel('max')
778
axes2.plot(numpy.max(data, axis=0))
779

780
axes3.set_ylabel('min')
781
axes3.plot(numpy.min(data, axis=0))
782
783

fig.tight_layout()
784

785
matplotlib.pyplot.show()
Greg Wilson's avatar
Greg Wilson committed
786
~~~
787
{: .python}
788

789
![The Previous Plots as Subplots]({{ site.github.url }}/fig/01-numpy_80_0.png)
790

791
The [call](reference.html#function-call) to `loadtxt` reads our data,
792
793
and the rest of the program tells the plotting library
how large we want the figure to be,
794
that we're creating three subplots,
795
796
797
what to draw for each one,
and that we want a tight layout.
(Perversely,
798
if we leave out that call to `fig.tight_layout()`,
799
the graphs will actually be squeezed together more closely.)
800

801
> ## Scientists Dislike Typing
802
803
804
805
806
807
808
>
> We will always use the syntax `import numpy` to import NumPy.
> However, in order to save typing, it is
> [often suggested](http://www.scipy.org/getting-started.html#an-example-script)
> to make a shortcut like so: `import numpy as np`.
> If you ever see Python code online using a NumPy function with `np`
> (for example, `np.loadtxt(...)`), it's because they've used this shortcut.
809
{: .callout}
810

811
> ## Check Your Understanding
mboisson's avatar
mboisson committed
812
813
814
>
> Draw diagrams showing what variables refer to what values after each statement in the following program:
>