From 06a42e2af40697bd2c95843aee1b75bc51d4270d Mon Sep 17 00:00:00 2001
From: Alfredo Canziani <alfredo.canziani@gmail.com>
Date: Wed, 29 Jun 2016 22:47:50 -0400
Subject: Add optim.Logger() documentation

---
 README.md           |   2 +-
 doc/logger.md       |  73 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 doc/logger_plot.png | Bin 0 -> 45532 bytes
 3 files changed, 74 insertions(+), 1 deletion(-)
 create mode 100644 doc/logger.md
 create mode 100644 doc/logger_plot.png

diff --git a/README.md b/README.md
index 716e08f..561621b 100644
--- a/README.md
+++ b/README.md
@@ -5,4 +5,4 @@ This package contains several optimization routines and a logger for [Torch](htt
 
  * [Overview](doc/intro.md);
  * [Optimization algorithms](doc/algos.md);
- * Logger.
+ * [Logger](doc/logger.md).
diff --git a/doc/logger.md b/doc/logger.md
new file mode 100644
index 0000000..b7797d2
--- /dev/null
+++ b/doc/logger.md
@@ -0,0 +1,73 @@
+<a name='optim.logger'></a>
+# Logger
+
+`optim` provides also logging and live plotting capabilities via the `optim.Logger()` function.
+
+Live logging is essential to monitor the *network accuracy* and *cost function* during training and testing, for spotting *under-* and *over-fitting*, for *early stopping* or just for monitoring the health of the current optimisation task.
+
+
+## Logging data
+
+Let walk through an example to see how it works.
+
+We start with initialising our logger connected to a text file `accuracy.log`.
+
+```lua
+logger = optim.Logger('accuracy.log')
+```
+
+We can decide to log on it, for example, *training* and *testing accuracies*.
+
+```lua
+logger:setNames{'Training acc.', 'Test acc.'}
+```
+
+And now we can populate our logger randomly.
+
+```lua
+for i = 1, 10 do
+   trainAcc = math.random(0, 100)
+   testAcc = math.random(0, 100)
+   logger:add{trainAcc, testAcc}
+end
+```
+
+We can `cat` `accuracy.log` and see what's in it.
+
+```
+Training acc.   Test acc.
+ 7.0000e+01      5.9000e+01
+ 7.6000e+01      8.0000e+00
+ 6.6000e+01      3.4000e+01
+ 7.4000e+01      4.3000e+01
+ 5.7000e+01      1.1000e+01
+ 5.0000e+00      9.8000e+01
+ 7.1000e+01      1.7000e+01
+ 9.8000e+01      2.7000e+01
+ 3.5000e+01      4.7000e+01
+ 6.8000e+01      5.8000e+01
+```
+
+## Visualising logs
+
+OK, cool, but how can we actually see what's going on?
+
+To have a better grasp of what's happening, we can plot our curves.
+We need first to specify the plotting style, choosing from:
+
+  * `.` for dots
+  * `+` for points
+  * `-` for lines
+  * `+-` for points and lines
+  * `~` for using smoothed lines with cubic interpolation
+  * `|` for using boxes
+  * custom string, one can also pass custom strings to use full capability of gnuplot.
+
+```lua
+logger:style{'+-', '+-'}
+logger:plot()
+```
+
+![Logging plot](logger_plot.png)
+
+If we'd like an interactive visualisation, we can put the `logger:plot()` instruction within the `for` loop, and the chart will be updated at every iteration.
diff --git a/doc/logger_plot.png b/doc/logger_plot.png
new file mode 100644
index 0000000..c5e86ae
Binary files /dev/null and b/doc/logger_plot.png differ
-- 
cgit v1.2.3