Try a simple test via GitHub Actions

.github/workflows/ci.yml

@@ -0,0 +1,45 @@
+name: CI
+on: [push, pull_request]
+jobs:
+  pre_job:
+    name: Check duplicate
+    runs-on: ubuntu-latest
+    outputs:
+      should_skip: ${{ steps.skip_check.outputs.should_skip }}
+    steps:
+      - id: skip_check
+        uses: fkirc/skip-duplicate-actions@12aca0a884f6137d619d6a8a09fcc3406ced5281
+        with:
+          cancel_others: true
+          concurrent_skipping: same_content
+          do_not_skip: '["pull_request", "schedule", "workflow_dispatch"]'
+
+  build:
+    name: Test
+    needs: pre_job
+    if: ${{ needs.pre_job.outputs.should_skip != 'true' }}
+    runs-on: ubuntu-latest
+    steps:
+      - run: git config --global core.autocrlf input
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+      - uses: actions/setup-python@v4
+        with:
+          python-version: '3.11'
+          cache: 'pipenv'
+      - name: Install pipenv
+        run: curl https://raw.githubusercontent.com/pypa/pipenv/master/get-pipenv.py | python
+      - run: pipenv install
+      - name: Compile database
+        run: pipenv run python compile_db.py --output test/hs.db < test/patterns.json > test/patterns_final.json
+      - name: Generate test data
+        run: pipenv run python gen_test_data.py
+      - name: Preprocess training data
+        run: pipenv run python preprocess.py train --database test/hs.db --sherlock-path test/ --output-dir test/
+      - name: Train the model
+        run: pipenv run python train.py --sherlock-path test/ --input-dir test/ --output-dir test/
+      - name: Preprocess test data
+        run: pipenv run python preprocess.py test --database test/hs.db --sherlock-path test/ --output-dir test/
+      - name: Evaluate the model
+        run: pipenv run python test.py --sherlock-path test/ --input-dir test/ | grep "weighted avg       1.00"

.gitignore

@@ -10,4 +10,5 @@ patterns.json
 *.json
 *.h5
 *.png
+*.parquet
 regex101/

compile_db.py

@@ -1,10 +1,15 @@
+import argparse
 import ast
 import hyperscan
 import json
 import pickle
 import sys
 
 
+parser = argparse.ArgumentParser()
+parser.add_argument("-o", "--output", default="hs.db")
+args = parser.parse_args()
+
 sys.stderr.write("Collecting patterns...\n")
 regexes = set()
 for line in sys.stdin:
@@ -34,5 +39,5 @@
 # Compile the final database and save to file
 sys.stderr.write("Compiling %d patterns...\n" % len(patterns))
 db.compile(expressions=patterns, ids=ids, flags=flags)
-with open("hs.db", "wb") as f:
+with open(args.output, "wb") as f:
     pickle.dump([len(patterns), hyperscan.dumpb(db)], f)

gen_test_data.py

@@ -0,0 +1,16 @@
+import pandas as pd
+
+
+data = ["['a', 'b', 'c']", "['1', '2', '3']"]
+df = pd.DataFrame(data, columns=['values'])
+df.to_parquet('test/test_values.parquet', index=True)
+
+df = pd.DataFrame(data * 100, columns=['values'])
+df.to_parquet('test/train_values.parquet', index=True)
+
+labels = ["alpha", "numeric"]
+df = pd.DataFrame(labels, columns=['type'])
+df.to_parquet('test/test_labels.parquet', index=True)
+
+df = pd.DataFrame(labels * 100, columns=['type'])
+df.to_parquet('test/train_labels.parquet', index=True)

preprocess.py

@@ -23,13 +23,16 @@
 
 parser = argparse.ArgumentParser()
 parser.add_argument("dataset", choices=["train", "test"])
+parser.add_argument("--database", default="hs.db")
+parser.add_argument("--sherlock-path", default="../sherlock-project/data/data/raw")
+parser.add_argument("--output-dir", default=".")
 args = parser.parse_args()
 
-output_file = f"preprocessed_{args.dataset}.txt"
+output_file = os.path.join(args.output_dir, f"preprocessed_{args.dataset}.txt")
 
 # Load the precompiled regular expression database
 sys.stderr.write("Loading regexes from file…\n")
-with open("hs.db", "rb") as f:
+with open(args.database, "rb") as f:
     [num_patterns, bdb] = pickle.load(f)
     db = hyperscan.loadb(bdb)
     # Scratch is not correctly initialized for deserialized databses
@@ -43,7 +46,7 @@ def on_match(match_id, from_idx, to_idx, flags, context):
 
 
 # Load the values
-pq_values = ParquetFile(f"../sherlock-project/data/data/raw/{args.dataset}_values.parquet")
+pq_values = ParquetFile(os.path.join(args.sherlock_path, f"{args.dataset}_values.parquet"))
 
 # Remove the output if it exists
 if os.path.exists(output_file):

test.py

@@ -1,3 +1,5 @@
+import argparse
+import os
 import sys
 
 import numpy as np
@@ -10,22 +12,27 @@
 
 BATCH_SIZE = 1000
 
+parser = argparse.ArgumentParser()
+parser.add_argument("--sherlock-path", default="../sherlock-project/data/data/raw")
+parser.add_argument("--input-dir", default=".")
+args = parser.parse_args()
+
 sys.stderr.write("Loading labels...\n")
-pq_labels = ParquetFile("../sherlock-project/data/data/raw/test_labels.parquet")
+pq_labels = ParquetFile(os.path.join(args.sherlock_path, "test_labels.parquet"))
 labels = pd.DataFrame(
     {"type": pd.Categorical(pq_labels.read(columns=["type"]).columns[0].to_numpy())}
 )
 le = LabelEncoder()
-le.classes_ = np.load("classes.npy", allow_pickle=True)
+le.classes_ = np.load(os.path.join(args.input_dir, "classes.npy"), allow_pickle=True)
 # labels = le.transform(labels.values.ravel())
 num_examples = len(labels)
 
-model = model_from_json(open("nn_model_sherlock.json", "r").read())
-model.load_weights("nn_model_weights_sherlock.h5")
+model = model_from_json(open(os.path.join(args.input_dir, "nn_model_sherlock.json"), "r").read())
+model.load_weights(os.path.join(args.input_dir, "nn_model_weights_sherlock.h5"))
 
 sys.stderr.write("Evaluating...\n")
 labels_pred = [""] * len(labels)
-preprocessed = open("preprocessed_test.txt", "r")
+preprocessed = open(os.path.join(args.input_dir, "preprocessed_test.txt"), "r")
 batch = 0
 with tqdm(total=len(labels)) as pbar:
     while True:

test/.gitignore

@@ -0,0 +1 @@
+!patterns.json

test/patterns.json

@@ -0,0 +1,2 @@
+"^[A-Za-z]$"
+"^[0-9]$"

train.py

@@ -1,3 +1,5 @@
+import argparse
+import os
 import sys
 
 import numpy as np
@@ -17,8 +19,14 @@
 
 BATCH_SIZE = 1000
 
+parser = argparse.ArgumentParser()
+parser.add_argument("--sherlock-path", default="../sherlock-project/data/data/raw")
+parser.add_argument("--input-dir", default=".")
+parser.add_argument("--output-dir", default=".")
+args = parser.parse_args()
+
 sys.stderr.write("Loading labels...\n")
-pq_labels = ParquetFile("../sherlock-project/data/data/raw/train_labels.parquet")
+pq_labels = ParquetFile(os.path.join(args.sherlock_path, "train_labels.parquet"))
 labels = pd.DataFrame(
     {
         "type": pd.Categorical(
@@ -31,10 +39,10 @@
 # Encode the labels as integers
 le = LabelEncoder().fit(labels.values.ravel())
 labels = le.transform(labels.values.ravel())
-np.save("classes.npy", le.classes_)
+np.save(os.path.join(args.output_dir, "classes.npy"), le.classes_)
 
 # Load one row just to get the shape of the input
-preprocessed = open("preprocessed_train.txt", "r")
+preprocessed = open(os.path.join(args.input_dir, "preprocessed_train.txt"), "r")
 matrix = np.loadtxt(preprocessed, max_rows=1)
 regex_shape = matrix.shape[0]
 
@@ -78,9 +86,9 @@
     loss="categorical_crossentropy",
     metrics=["categorical_accuracy"],
 )
-open("nn_model_sherlock.json", "w").write(model.to_json())
+open(os.path.join(args.output_dir, "nn_model_sherlock.json"), "w").write(model.to_json())
 
-preprocessed = open("preprocessed_train.txt", "r")
+preprocessed = open(os.path.join(args.input_dir, "preprocessed_train.txt"), "r")
 i = 0
 with tqdm(total=len(labels)) as pbar:
     while True:
@@ -105,4 +113,4 @@
         pbar.update(len(matrix))
 
 # Save the trained model weights
-model.save_weights("nn_model_weights_sherlock.h5")
+model.save_weights(os.path.join(args.output_dir, "nn_model_weights_sherlock.h5"))