Add CSV-loading mode #1597
Add CSV-loading mode #1597
Conversation
|
I will get this reviewed by Tolga once he is added to tensorboard-team, but feel free to take a look as well Stephan |
|
@tolga-b Please review |
| @@ -1476,8 +1479,13 @@ <h2>Create a distance feature</h2> | |||
| const inferenceErrorStr = ' Inference error'; | |||
| const inferenceAbsErrorStr = ' Inference absolute error'; | |||
| const inferenceSquaredErrorStr = ' Inference squared error'; | |||
| const inferenceScoreStr = ' Inference score '; | |||
| const inferenceScoreStr = ' Inference score'; | |||
There was a problem hiding this comment.
space in the end is intentional since we append model number directly to inferenceScoreStr when showing inference scores for multiple models now.
| @@ -477,6 +493,11 @@ Polymer({ | |||
| } | |||
| min = Math.min(0, min) * clipSaliencyRatio; | |||
| max = Math.max(0, max) * clipSaliencyRatio; | |||
| if (min < 0 && max > Math.abs(min)) { | |||
There was a problem hiding this comment.
Could you comment on this change, why are we making values symmetric around zero?
| from glob import glob | ||
| from grpc.beta import implementations | ||
| import random | ||
| import numpy as np |
There was a problem hiding this comment.
good catch. not anymore. removed from here and build file
| @@ -274,6 +276,20 @@ Polymer({ | |||
| .getFeatureLists()!.getFeatureListMap(); | |||
| }, | |||
|
|
|||
| setFilteredFeaturesList: function(featureList: NameAndFeature[], | |||
| searchValue: string, saliency: SaliencyMap) { | |||
| this.filteredFeaturesList = []; | |||
There was a problem hiding this comment.
JS style nit: please -2 indent in this function block and the one below
| // for negative and positive attributions are the same, for visual | ||
| // consistency. | ||
| if (min < 0 && max > Math.abs(min)) { | ||
| min = -1 * max; |
There was a problem hiding this comment.
another style nit: -1 indent before max
| @@ -1071,7 +1074,7 @@ Polymer({ | |||
| * be used in css classes/ids. | |||
| */ | |||
| sanitizeFeature: function(feat: string) { | |||
| let sanitized = feat; | |||
| let sanitized = feat.trim(); | |||
There was a problem hiding this comment.
nit: indentation looks off
There was a problem hiding this comment.
Just saying, because of L1079, sanitized may still have the whitespace. Perhaps use sanitized in L1078 and L1079?
| compare-title="[[compareTitle]]"> | ||
| compare-title="[[compareTitle]]" | ||
| saliency="[[attribution]]" | ||
| show-saliency="true" |
There was a problem hiding this comment.
you can omit ="true" when specifying boolean prop.
| @@ -2829,6 +2849,70 @@ <h2>Create a distance feature</h2> | |||
| this.selectedExampleAndInference = | |||
| this.selected.length > 0 | |||
| ? this.examplesAndInferences[this.selected[0]] : null; | |||
| requestAnimationFrame(()=> this.setAttributions()); | |||
There was a problem hiding this comment.
for legibility, can you put one whitespace before =>? thanks!
also, often, if you start a RAF, it is a good practice to cancel it on unmount or detached. Please consider cancelling the RAF.
| @@ -2093,6 +2108,11 @@ <h2>Create a distance feature</h2> | |||
| this.isBinaryClassification_(modelType, multiClass); | |||
| }, | |||
|
|
|||
| enablePartialDependencePlots_: function(selectedExampleAndInference, modelName, inferenceAddress) { | |||
There was a problem hiding this comment.
like areExamplesEditable_ method, it might read better if this method name is isPartialDependencePlotsEnabled_ or something like this
| */ | ||
| setAttributions: function() { | ||
| const attrib = {}; | ||
| const selectedData = this.visdata[this.selected[0]]; |
There was a problem hiding this comment.
stupid question: couldn't read the entire file but it roughly looks like only the first element in selected is used. Why is it using array for its type?
There was a problem hiding this comment.
Can you please provide what kind of structure visdata has in the code? thanks!
There was a problem hiding this comment.
selected is bound to facets-dive selected attribute, which is an array of indices of selected elements in facets-dive. in the what-if tool we only make use of the first selected element.
added more comments to visdata and selected definitions to specify structure.
If given a CSV file (with .csv extension) as the examples path, then the tool will create and visualize examples from CSV as opposed to tfrecords. Additionally, the CSV can contain columns for model inference output and per-feature attribution for a given inference - as might be provided by the TF model analysis project (TFMA) project. The attributions are visualized in the example viewer per code that already exists in that component.
In CSV mode, examples aren't editable and PD plots aren't accessible since there is no live model to infer examples against.