Assistant Professor in the UW Allen School and Co-director of the Interactive Data Lab.
http://idl.cs.washington.edu/
Pyxis: Analytic Knowledge
import { aq, op } from "@uwdata/arquero";
desc = aq.desc;
baltimoreCrimeJson = (await fetch("https://raw.githubusercontent.com/leibatt/pyxis/main/datasets/BPD_Part_1_Victim_Based_Crime_Data2.json")).json();
baltimoreCrime = pyxis.jsonObjectToDataset(baltimoreCrimeJson,"baltimoreCrime");
baltimoreCrime.records[190000];
peakCrimesTransformation = {
let t = {
sources: [baltimoreCrime],
ops: ["groupby","rollup","orderby","filter"], // list all verbs for our records
transforms: [
{
op: "groupby",
args: ["CrimeDate"]
},
{
op: "rollup",
args: [{
count: op.count()
}]
},
{
op: "orderby",
args: [desc("count")]
},
{
op: "filter",
args: [() => op.rank() <= 2]
}
]
};
return t;
};
peakCrimesResults = pyxis.transformation.arquero.executeDataTransformation(peakCrimesTransformation);
peakCrimesNode = new pyxis.AnalyticKnowledgeNode(
"peakCrimes", // node name
Date.now(), // timestamp
peakCrimesTransformation, // transformation
null, // relationshipModel
() => pyxis.transformation.arquero.executeDataTransformation(peakCrimesTransformation) // results
);
peakCrimesNode.name;
The second parameter is a timestamp representing when the knowledge was recorded.
peakCrimesNode.timestamp;
The third parameter is the Pyxis DataTransformation object associated with the knowledge node. This parameter is arguably the most important one, since it captures the analysis findings we want to record.
peakCrimesNode.transformation;
peakCrimesNode.relationshipModel;
peakCrimesNode.results();
crimeLocationModel = new pyxis.DecisionTreeClassificationRelationshipModel(
"baltimoreCrimes", // give the dataset a name, we can just call it baltimoreCrimes
[ // input attributes, the attributes used to predict a certain outcome
{
name: "Inside/Outside", // Attribute name from the baltimoreCrimes dataset
attributeType: pyxis.AttributeType.nominal // the type of attribute (quantitative, ordinal, or nominal)
},
{
name: "Premise",
attributeType: pyxis.AttributeType.nominal
}
],
{ // output attribute, the outcome to be predicted
name: "Description",
attributeType: pyxis.AttributeType.nominal
}
);
crimeLocationModel.train(baltimoreCrime);
crimeLocationNode = new pyxis.AnalyticKnowledgeNode(
"crimeLocation", // node name
Date.now(), // timestamp
null, // transformation
crimeLocationModel, // relationshipModel
() => baltimoreCrime // results
);
crimeLocationNode.relationshipModel;
crimeLocationNode.results();
crimeDistributionTransformation = {
let t = {
sources: [baltimoreCrime],
ops: ["filter","groupby","rollup"], // list all verbs for our records
transforms: [
{
op: "filter",
args: [(d) => d["CrimeDate"] === "04/27/2015"]
// op.year(d["CrimeDate"]) === 2015 && op.month(d["CrimeDate"]) === 3 && op.date(d["CrimeDate"]) === 27
},
{
op: "groupby",
args: ["Description"]
},
{
op: "rollup",
args: [{
count: op.count()
}]
},
{
op: "orderby",
args: [desc("count")]
}
]
};
return t;
};
crimeDistributionNode = new pyxis.AnalyticKnowledgeNode(
"crimeDistribution", // name
Date.now(), // timestamp
crimeDistributionTransformation, // transformation
null, // relationshipModel
() => pyxis.transformation.arquero.executeDataTransformation(crimeDistributionTransformation), // results
"Distribution of crimes observed for the April 27, 2015 peak. Burglaries are the most frequent crime type on this day (210 crimes)." // description
);
crimeDistributionNode.results();
crimeDistributionNode.addSource(peakCrimesNode);
the addSource function adds the specified node to the list of sources for the target node and the list of nodes related to this node:
crimeDistributionNode.source;
crimeDistributionNode.related;
peakCrimesNode.target;
peakCrimesNode.related;
crimeLocationNode.addRelated(crimeDistributionNode);
crimeLocationNode.related;
crimeDistributionNode.related;
Purpose-built for displays of data
Observable is your go-to platform for exploring data and creating expressive data visualizations. Use reactive JavaScript notebooks for prototyping and a collaborative canvas for visual data exploration and dashboard creation.
