feature_vector_vis

Module: missions.nodes.visualization.feature_vector_vis

Visualize FeatureVector elements

Inheritance diagram for pySPACE.missions.nodes.visualization.feature_vector_vis:

Inheritance diagram of pySPACE.missions.nodes.visualization.feature_vector_vis

Class Summary

LLEVisNode([neighbors]) Show a 2d scatter plot of all FeatureVector based on Locally Linear Embedding (LLE) from MDP
MnistVizNode([mode, history_index, max_samples]) Node for plotting MNIST Data

Classes

LLEVisNode

class pySPACE.missions.nodes.visualization.feature_vector_vis.LLEVisNode(neighbors=15, **kwargs)[source]

Bases: pySPACE.missions.nodes.base_node.BaseNode

Show a 2d scatter plot of all FeatureVector based on Locally Linear Embedding (LLE) from MDP

This node collects all training examples it obtains along with their label. It computes than an embedding of all these examples in a 2d space using the “Locally Linear Embedding” algorithm and plots a scatter plot of the examples in this space.

Parameters

neighbors:

The number of neighbor vectors that should be considered for each instance during locally linear embedding

(optional, default: 15)

Exemplary Call

- 
    node : Time_Series_Source
-
    node : All_Train_Splitter
-
    node : Time_Domain_Features
-
    node : LLE_Vis
    parameters :
        neighbors : 10
-
    node : Nil_Sink

Known Issues: This node will use pylab.show() to show the figure. There is no store method implemented yet. On Macs, pylab.show() might sometimes fail due to a wrong plotting backend. A possible workaround in that case is to manually set the plotting backend to ‘MacOSX’. This has to be done before pylab is imported, so one can temporarily add “import matplotlib; matplotlib.use(‘MacOSX’)” to the very beginning of launch.py.

Author:

Jan Hendrik Metzen (jhm@informatik.uni-bremen.de)

Created:

2009/07/07

POSSIBLE NODE NAMES:
 
  • LLE_Vis
  • LLEVis
  • LLEVisNode
POSSIBLE INPUT TYPES:
 
  • FeatureVector

Class Components Summary

_execute(data)
_get_train_set(use_test_data) Returns the data that can be used for training
_stop_training([debug]) Stops the training, i.e.
_train(data, label) This node is not really trained but uses the labeled examples to generate a scatter plot.
input_types
is_supervised() Returns whether this node requires supervised training
is_trainable() Returns whether this node is trainable.
input_types = ['FeatureVector']
__init__(neighbors=15, **kwargs)[source]
is_trainable()[source]

Returns whether this node is trainable.

is_supervised()[source]

Returns whether this node requires supervised training

_get_train_set(use_test_data)[source]

Returns the data that can be used for training

_train(data, label)[source]

This node is not really trained but uses the labeled examples to generate a scatter plot.

_stop_training(debug=False)[source]

Stops the training, i.e. create the 2d representation

Uses the Locally Linear Embedding algorithm to create a 2d representation of the data and creates a 2d scatter plot.

_execute(data)[source]

MnistVizNode

class pySPACE.missions.nodes.visualization.feature_vector_vis.MnistVizNode(mode=None, history_index=0, max_samples=10, **kwargs)[source]

Bases: pySPACE.missions.nodes.base_node.BaseNode

Node for plotting MNIST Data

Parameters
mode:

One of FeatureVector, PredictionVector, and nonlinear.

If FeatureVector is taken, the data is assumed to be in the 28x28 format and can be visualized like the original data.

If PredictionVector is chosen, the affine backtransformation approach is used. If possible, the visualization is enhanced by the average data found in the data history at the history_index.

If nonlinear os used, a nonlinear processing chain is assumed for calculating the backtransformation with derivatives with the sample at the

If not specified, the input data type is used.

(recommended, default: input type)

history_index:

Index for determining the averaging data or the data for calculating the derivative from prediction vectors. To save the respective data, the keep_in_history parameter has to be used, in the node, which produces the needed data. This can be a Noop node at the beginning.

By default the last stored sample is used.

(recommended, default: last sample)

max_samples:

In case of the nonlinear mode, a backtransformation graphic must be generated for every data sample. To reduce memory usage, only the first max_*samples training samples are used.

(optional, default: 10)

Exemplary Call

- node : MnistViz
POSSIBLE NODE NAMES:
 
  • MnistViz
  • MnistVizNode
POSSIBLE INPUT TYPES:
 
  • PredictionVector
  • FeatureVector
  • TimeSeries

Class Components Summary

_train(data, label) Average data with labels (no real training)
input_types
is_supervised() Labels are required for visualization
is_trainable() Labels are required for visualization
store_state(result_dir[, index]) Main method which generates and stores the graphics
__init__(mode=None, history_index=0, max_samples=10, **kwargs)[source]
_train(data, label)[source]

Average data with labels (no real training)

store_state(result_dir, index=None)[source]

Main method which generates and stores the graphics

is_trainable()[source]

Labels are required for visualization

is_supervised()[source]

Labels are required for visualization

input_types = ['PredictionVector', 'FeatureVector', 'TimeSeries']