Tutorial 4 - Data extraction with defects
This tutoral is a continuation of the Tutorial 3.
This tutorial will demonstrate how to use Archery to extract data from one Excel spreadsheet with defects and its tagging capabilities. Tagging enable to fix a schema for the extracted data and ease the loading into a database for example. To demonstrate the usage of this framework, we will load a document with a somewhat complex layout, as seen here:
Setup Archery
Import the packages and setup the main class:
package com.github.romualdrousseau.archery.examples;
import java.util.EnumSet;
import java.util.List;
import com.github.romualdrousseau.archery.Document;
import com.github.romualdrousseau.archery.DocumentFactory;
import com.github.romualdrousseau.archery.parser.LayexTableParser;
public class Tutorial4 implements Runnable {
public Tutorial4() {
}
@Override
public void run() {
// Code will come here
}
public static void main(final String[] args) {
new Tutorial4().run();
}
}
pom.xml
Archery has a very modular design where each functionality can be loaded separatly. We add the "archery-net-classifier" module to enable the tagging capabilities. This module use TensorFlow for Java. The following depedencies are required to run the code of this tutorial:
<!-- Archery Framework -->
<dependency>
<groupId>com.github.romualdrousseau</groupId>
<artifactId>archery</artifactId>
<version>${archery.version}</version>
</dependency>
<dependency>
<groupId>com.github.romualdrousseau</groupId>
<artifactId>archery-layex-parser</artifactId>
<version>${archery.version}</version>
</dependency>
<dependency>
<groupId>com.github.romualdrousseau</groupId>
<artifactId>archery-net-classifier</artifactId>
<version>${archery.version}</version>
</dependency>
<dependency>
<groupId>com.github.romualdrousseau</groupId>
<artifactId>archery-csv</artifactId>
<version>${archery.version}</version>
</dependency>
<dependency>
<groupId>com.github.romualdrousseau</groupId>
<artifactId>archery-excel</artifactId>
<version>${archery.version}</version>
</dependency>
Load base model
To parse a document, Archery needs a model that will contains the parameters required to the parsing. Instead to start from an empty Model (See Tutorial 10), we will start from an existing one and we will adapt it for our document. You can find a list and details of all models here.
The base model, we will use, is "sales-english" that has been trained on 200+ english documents containing distributor data and with a large range of different layouts.
Because we use the tagging capabilities in this tutorial, here are a subset of tags recognized by the base model:
[
{
"name" : "tags",
"doc" : "Tags recognized by sales-english model.",
"settings" : {
"types" : [
"none",
"date",
"dateYear",
"dateMonth",
"wholesalerCode",
"wholesalerName",
"customerCode",
"customerName",
"customerType",
"customerGroup",
"country",
"postalCode",
"adminArea1",
"adminArea2",
"adminArea3",
"adminArea4",
"locality",
"address",
"productCode",
"productName",
"amount",
"unitPrice",
"quantity",
"bonusQuantity",
"returnQuantity",
"totalQuantity",
"billToCode",
"billToName",
"transactionType",
"invoiceNumber",
"invoiceLineNumber",
"batchNumber",
"expiryDate",
"creditReasonCode",
"requesterName"
],
"requiredTags" : [
"quantity",
"productCode,productName"
]
}
}
]
The base model already recognize some entities such as DATE and NUMBER. We will setup the model to add one new entity PRODUCTNAME and we will configure a layex to extract the different elements of the documents. You can find more details about layex here.
final var model = Common.loadModelFromGitHub("sales-english");
// Add product name entity to the model
model.getEntityList().add("PRODUCTNAME");
model.getPatternMap().put("\\D+\\dml", "PRODUCTNAME");
model.update();
// Add a layex to the model
final var tableParser = new LayexTableParser(
List.of("(v.$)+"),
List.of("(()(S+$))(()([/^TOTAL/|v].+$)())+(/TOTAL/.+$)"));
model.registerTableParser(tableParser);
Load the document
We load the document by creating a document instance with the model. The hint "Document.Hint.INTELLI_LAYOUT" tell the document instance that the document has a complex layout. We also add the hint "Document.Hint.INTELLI_TAG" to tell that the tabular result must be tagged. The recipe "sheet.setCapillarityThreshold(0)" tell the parser engine to extract the features as small as possible:
final var file = Common.loadData("document with defect.xlsx", this.getClass());
try (final var doc = DocumentFactory.createInstance(file, "UTF-8")
.setModel(model)
.setHints(EnumSet.of(Document.Hint.INTELLI_LAYOUT, Document.Hint.INTELLI_TAG))
.setRecipe("sheet.setCapillarityThreshold(0)")) {
...
}
Output the tabular result
Finally, we iterate over the sheets, rows and cells and output the data on the console:
doc.sheets().forEach(s -> Common.addSheetDebugger(s).getTable().ifPresent(t -> {
Common.printTags(t.headers());
Common.printRows(t.rows());
}));
Note that now we are printing the tags of the headers and not their names.
2024-03-09 23:54:59 INFO Common:37 - Loaded resource: /models/sales-english.json
2024-03-09 23:54:59 INFO Common:37 - Loaded resource: /data/document with defect.xlsx
2024-03-09 23:55:02 DEBUG Common:64 - Extracting features ...
2024-03-09 23:55:02 DEBUG Common:68 - Generating Layout Graph ...
2024-03-09 23:55:02 DEBUG Common:72 - Assembling Tabular Output ...
============================== DUMP GRAPH ===============================
Sheet1
|- A document very important DATE META(1, 1, 4, 1, 1, 1)
|- |- PRODUCTNAME META(1, 4, 1, 4, 1, 1)
|- |- |- Date Client Qty Amount DATA(1, 5, 4, 10, 6, 4) (1)
|- |- PRODUCTNAME META(1, 11, 1, 11, 1, 1)
|- |- |- Date Client Qty Amount DATA(1, 12, 4, 17, 6, 4) (2)
|- |- PRODUCTNAME META(1, 19, 1, 19, 1, 1)
|- |- |- Date Client Qty Amount DATA(1, 20, 4, 25, 6, 4) (3)
================================== END ==================================
2024-03-09 23:55:03.459511: I external/org_tensorflow/tensorflow/cc/saved_model/reader.cc:45] Reading SavedModel from: /tmp/model-9696004103989867291
2024-03-09 23:55:03.461712: I external/org_tensorflow/tensorflow/cc/saved_model/reader.cc:89] Reading meta graph with tags { serve }
2024-03-09 23:55:03.461749: I external/org_tensorflow/tensorflow/cc/saved_model/reader.cc:130] Reading SavedModel debug info (if present) from: /tmp/model-9696004103989867291
2024-03-09 23:55:03.461804: I external/org_tensorflow/tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-03-09 23:55:03.477397: I external/org_tensorflow/tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:354] MLIR V1 optimization pass is not enabled
2024-03-09 23:55:03.478886: I external/org_tensorflow/tensorflow/cc/saved_model/loader.cc:229] Restoring SavedModel bundle.
2024-03-09 23:55:03.537380: I external/org_tensorflow/tensorflow/cc/saved_model/loader.cc:213] Running initialization op on SavedModel bundle at path: /tmp/model-9696004103989867291
2024-03-09 23:55:03.550411: I external/org_tensorflow/tensorflow/cc/saved_model/loader.cc:305] SavedModel load for tags { serve }; Status: success: OK. Took 90916 microseconds.
2024-03-09 23:55:03 DEBUG Common:77 - Done.
none date productName customerName quantity amount
A document very 2023-02-01 Product 1ml AAA 1 100
A document very 2023-02-01 Product 1ml BBB 1 100
A document very 2023-02-01 Product 1ml BBB 3 300
A document very 2023-02-01 Product 1ml AAA 1 100
A document very 2023-02-01 Product 2ml AAA 1 100
A document very 2023-02-01 Product 2ml BBB 2 200
A document very 2023-02-01 Product 2ml CCC 4 400
A document very 2023-02-01 Product 2ml DDD 1 100
A document very 2023-02-01 Product 3ml AAA 1 100
A document very 2023-02-01 Product 3ml CCC 1 100
A document very 2023-02-01 Product 3ml AAA 1 100
A document very 2023-02-01 Product 3ml DDD 1 100
On this output, we print out the graph of the document built during the parsing and we can see clearly the relation between the elements of the spreadsheet and how there are structured in tabular form. Observe how the column names have been replaced by tags describing the recognized columns.
Conclusion
Congratulations! You have loaded documents using Archery.
For more examples of using Archery, check out the tutorials.