diff --git a/index.html b/index.html index 62572e22..5f693c19 100644 --- a/index.html +++ b/index.html @@ -33,6 +33,7 @@ mkMenu("gismo", "GISMO"); mkMenu("manu", "MANUAL"); mkMenu("tuts", "TUTORIAL"); + // mkMenu("plumed_tuts", "PLUMED"); diff --git a/pages/plumed_tuts.html b/pages/plumed_tuts.html new file mode 100644 index 00000000..2d2a5a31 --- /dev/null +++ b/pages/plumed_tuts.html @@ -0,0 +1,21 @@ + + +
+ + +
diff --git a/pages/tuts/landmarks.html b/pages/tuts/landmarks.html index 3ad1772c..bcc17ea6 100644 --- a/pages/tuts/landmarks.html +++ b/pages/tuts/landmarks.html @@ -22,7 +22,6 @@

Selecting landmark points

space, dimlandmark also allows to assign weights proportional to the density of samples surrounding each sample.

-
running dimlandmark [+-] @@ -45,6 +44,100 @@

Selecting landmark points

- + +

+ Both of these strategies can also be run in PLUMED +

+ + +
+
+ using PLUMED for Random Selection and FPS [+-] +
+
+

+ The SelectRandomFrames allows user to select a small, representative subset of the high-dimensional data randomly, so that dimensionality reduction can be performed more effectively. USE_ALL_DATA can be replaced with RUN=ii (a number) to use datapoints only after ii steps while selecting randomly from the data. +

+
+ d: READ FILE=highd-dataset.in VALUES=c.*
+ OUTPUT_LANDMARKS ...
+ ARG=d.*
+ USE_ALL_DATA
+ METRIC=EUCLIDEAN
+ LANDMARKS={RANDOM N=300 SEED=-3241 }
+ LIST_FILE=landmark-random.out
+ ... OUTPUT_LANDMARKS
+
+

+ METRIC options specify the distance function employed, RANDOM keyword is used to specify the Random Sampling method to be employed and N is used for the number of landmarks to be selected. LIST_FILE is used to specify the output file where the landmarks are generated. SEED specifies the seed for the random number generator. It should be given different values for different runs. You can download the example file, the plumed input file and the script to prepare the input file as a tarball. +

+ +

+ The FarthestPointSampling allows user to select a small, representative subset of the high-dimensional data using Farthest Point Sampling technique. All the parameters remain same as defined in Random Sampling method except the key word FPS is used instead of RANDOM +

+
+ d: READ FILE=highd-dataset.in VALUES=c.*
+ OUTPUT_LANDMARKS ...
+ ARG=d.*
+ USE_ALL_DATA
+ METRIC=EUCLIDEAN
+ LANDMARKS={FPS N=300 SEED=1 }
+ LIST_FILE=landmark-fps.out
+ ... OUTPUT_LANDMARKS
+
+ +
+
+ +

+ Apart from the above mentioned strategies, there are some more strategies which can be used to select the landmarks when running the sketchmap algorithm with PLUMED. +

+ + + + +
+
+ using PLUMED for Selection with Stride and Staged Sampling [+-] +
+
+

+ The Stride Sampling allows user to select a representative subset of the high-dimensional data using strides of predefined constant step length. +

+
+ d: READ FILE=highd-dataset.in VALUES=c.*
+ OUTPUT_LANDMARKS ...
+ ARG=d.*
+ USE_ALL_DATA
+ METRIC=EUCLIDEAN
+ LANDMARKS={STRIDE N=300 }
+ LIST_FILE=landmark-stride.out
+ ... OUTPUT_LANDMARKS
+
+

+ +

+

+ The StagedSampling allows user to select a representative subset of the high-dimensional data using Staged Sampling technique. All the parameters remain same as defined in Random Sampling method except the key word STAGED is used and another parameter GAMMA is specified which determines TODO. +

+
+ d: READ FILE=highd-dataset.in VALUES=c.*
+ OUTPUT_LANDMARKS ...
+ ARG=d.*
+ USE_ALL_DATA
+ METRIC=EUCLIDEAN
+ LANDMARKS={STAGED N=300 SEED=1 GAMMA=0.5}
+ LIST_FILE=landmark-staged.out
+ ... OUTPUT_LANDMARKS
+
+

+ +

+
+
+ diff --git a/pages/tuts/sketch-map.html b/pages/tuts/sketch-map.html index 293e81a8..f060d193 100644 --- a/pages/tuts/sketch-map.html +++ b/pages/tuts/sketch-map.html @@ -47,12 +47,46 @@

Running sketch-map

+ +

+ The sketchmap can also be run under PLUMED framework. The following code explains the functionality available in the current implementation of the sketchmap algorithm in PLUMED. +

+
+
+ running Sketchmap in PLUMED [+-] +
+
+

+ Following code uses all dataset to run the sketch-map optimization which is decsribed in detail below. +

+
+ d: READ FILE=input.in VALUES=c.*
+ SKETCHMAP ...
+ ARG=d.*
+ NLOW_DIM=2
+ USE_ALL_DATA
+ HIGH_DIM_FUNCTION={SMAP R_0=1 A=3 B=2}
+ LOW_DIM_FUNCTION={SMAP R_0=1 A=1 B=2}
+ OUTPUT_FILE=smap-out
+ EMBEDDING_OFILE=smap-embed
+ ... SKETCHMAP
+ +
+ The parameters required and the functionality is detailed below: + + +
+
+
-
-
-
+

A detailed description of the process

@@ -110,8 +144,9 @@

Getting a starting map: running linear MDS

- - + + + Linear MDS embedding of 12-ala data, colored based on one of the dihedrals

Iterative metric MDS and pointwise-global optimization

diff --git a/sitemap.xml b/sitemap.xml index 4b9adacb..2f66fec6 100644 --- a/sitemap.xml +++ b/sitemap.xml @@ -13,6 +13,11 @@ http://epfl-cosmo.github.io/sketchmap/index.htm?page=tuts weekly 0.5000 + + + http://epfl-cosmo.github.io/sketchmap/index.htm?page=plumed_tuts + weekly + 0.5000 http://epfl-cosmo.github.io/sketchmap/index.htm?page=manu diff --git a/style.css b/style.css index d040dbc4..67bca531 100644 --- a/style.css +++ b/style.css @@ -130,5 +130,27 @@ div.code { padding:1em; font-family:"Courier"; font-weight:bold; background: #ffe4be; } +/* for plumed codes */ +div.pcbox { + padding:0; margin:2em; background: #fff5f1; clear:both; display:block +} + +div.pcbox .head { + background: #FFAC8A; font-weight:bold; + padding:5px; margin:0; cursor:pointer; + } + +div.pcbox .body { + padding:0.5em; margin:0; + } + +/* links & pc */ +a img { border:0 } + +.pcode {font-family:"Courier"; font-weight:bold;} +div.pcode { + padding:1em; font-family:"Courier"; font-weight:bold; + background: #ffe4be; +}