model: tensorflow hub: Add NLP model

* bert and nnlm embedding
* model: tensorflow: util: Make tensorflow config

Signed-off-by: John Andersen <[email protected]>

Author: 0dust

.ci/run.sh

@@ -20,6 +20,10 @@ function run_plugin() {
 
   "${PYTHON}" -m pip install -U "${SRC_ROOT}"
 
+  if [ "x${PLUGIN}" = "xmodel/tensorflow_hub" ]; then
+    "${PYTHON}" -m pip install -U "${SRC_ROOT}/model/tensorflow"
+  fi
+
   cd "${PLUGIN}"
   PACKAGE_NAME=$(dffml service dev setuppy kwarg name setup.py)
   "${PYTHON}" -m pip install -e .

.github/workflows/testing.yml

@@ -37,7 +37,7 @@ jobs:
       fail-fast: false
       max-parallel: 40
       matrix:
-        plugin: [., examples/shouldi, model/tensorflow, model/scratch, model/scikit, source/mysql, feature/git, feature/auth, service/http, config/yaml]
+        plugin: [., examples/shouldi, model/tensorflow, model/tensorflow_hub, model/scratch, model/scikit, source/mysql, feature/git, feature/auth, service/http, config/yaml]
         python-version: [3.7]
 
     steps:

CHANGELOG.md

@@ -6,6 +6,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 ### Added
+- Tensorflow hub NLP models.
 - Notes on development dependencies in `setup.py` files to codebase notes.
 
 ## [0.3.3] - 2020-02-10

dffml/service/dev.py

@@ -46,6 +46,7 @@
     ("model", "tensorflow"),
     ("model", "scratch"),
     ("model", "scikit"),
+    ("model", "tensorflow_hub"),
     ("examples", "shouldi"),
     ("feature", "git"),
     ("feature", "auth"),

docs/plugins/dffml_model.rst

@@ -297,6 +297,179 @@ predict).
   - default: [12, 40, 15]
   - List length is the number of hidden layers in the network. Each entry in the list is the number of nodes in that hidden layer
 
+dffml_model_tensorflow_hub
+--------------------------
+
+.. code-block:: console
+
+    pip install dffml-model-tensorflow-hub
+
+
+text_classifier
+~~~~~~~~~~~~~~~
+
+*Core*
+
+Implemented using Tensorflow hub pretrained models.
+
+.. code-block:: console
+
+    $ cat > train.csv << EOF
+    sentence,sentiment
+    Life is good,1
+    This book is amazing,1
+    It's a terrible movie,0
+    Global warming is bad,0
+    EOF
+    $ cat > test.csv << EOF
+    sentence,sentiment
+    I am not feeling good,0
+    Our trip was full of adventures,1
+    EOF
+    $ dffml train \
+        -model text_classifier \
+        -model-epochs 30 \
+        -model-predict sentiment:int:1 \
+        -model-classifications 0 1  \
+        -model-clstype int \
+        -sources f=csv \
+        -source-filename train.csv \
+        -model-features \
+          sentence:str:1 \
+        -model-model_path "https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim-with-oov/1" \
+        -model-add_layers \
+        -model-layers "Dense(units=512, activation='relu')" "Dense(units=2, activation='softmax')" \
+        -log debug
+    $ dffml accuracy \
+        -model text_classifier \
+        -model-predict sentiment:int:1 \
+        -model-classifications 0 1 \
+        -model-model_path "https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim-with-oov/1" \
+        -model-clstype int \
+        -sources f=csv \
+        -source-filename test.csv \
+        -model-features \
+          sentence:str:1 \
+        -log critical
+        1.0
+    $ dffml predict all \
+        -model text_classifier \
+        -model-predict sentiment:int:1 \
+        -model-classifications 0 1 \
+        -model-model_path "https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim-with-oov/1" \
+        -model-clstype int \
+        -sources f=csv \
+        -source-filename test.csv \
+        -model-features \
+          sentence:str:1 \
+        -log debug
+    [
+        {
+            "extra": {},
+            "features": {
+                "sentence": "I am not feeling good",
+                "sentiment": 0
+            },
+            "key": "0",
+            "last_updated": "2020-02-15T02:54:02Z",
+            "prediction": {
+                "sentiment": {
+                    "confidence": 0.7630850076675415,
+                    "value": 0
+                }
+            }
+        },
+        {
+            "extra": {},
+            "features": {
+                "sentence": "Our trip was full of adventures",
+                "sentiment": 1
+            },
+            "key": "1",
+            "last_updated": "2020-02-15T02:54:02Z",
+            "prediction": {
+                "sentiment": {
+                    "confidence": 0.6673157811164856,
+                    "value": 1
+                }
+            }
+        }
+    ]
+
+**Args**
+
+- predict: Feature
+
+  - Feature name holding classification value
+
+- classifications: List of strings
+
+  - Options for value of classification
+
+- features: List of features
+
+  - Features to train on
+
+- trainable: String
+
+  - default: True
+  - Tweak pretrained model by training again
+
+- batch_size: Integer
+
+  - default: 120
+  - Batch size
+
+- max_seq_length: Integer
+
+  - default: 256
+  - Length of sentence, used in preprocessing of input for bert embedding
+
+- add_layers: Boolean
+
+  - default: False
+  - Add layers on the top of pretrianed model/layer
+
+- embedType: String
+
+  - default: None
+  - Type of pretrained embedding model, required to be set to `bert` to use bert pretrained embedding
+
+- layers: List of strings
+
+  - default: None
+  - Extra layers to be added on top of pretrained model
+
+- model_path: String
+
+  - default: https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim-with-oov/1
+  - Pretrained model path/url
+
+- optimizer: String
+
+  - default: adam
+  - Optimizer used by model
+
+- metrics: String
+
+  - default: accuracy
+  - Metric used to evaluate model
+
+- clstype: Type
+
+  - default: <class 'str'>
+  - Data type of classifications values
+
+- epochs: Integer
+
+  - default: 10
+  - Number of iterations to pass over all repos in a source
+
+- directory: String
+
+  - default: /home/user/.cache/dffml/tensorflow_hub
+  - Directory where state should be saved
+
 dffml_model_scratch
 -------------------
 

model/scikit/tests/test_scikit_integration.py

@@ -9,14 +9,12 @@
 import contextlib
 
 import numpy as np
+from sklearn.datasets import make_blobs
 
 from dffml.cli.cli import CLI
 from dffml.util.asynctestcase import IntegrationCLITestCase
 
 
-from sklearn.datasets import make_blobs
-
-
 class TestScikitClassification(IntegrationCLITestCase):
     async def test_run(self):
         self.required_plugins("dffml-model-scikit")
@@ -306,7 +304,6 @@ async def test_run(self):
                 "training_data=csv",
                 "-source-filename",
                 train_file,
-                "-source-readonly",
             )
             # Assess accuracy
             await CLI.cli(
@@ -318,7 +315,6 @@ async def test_run(self):
                         *features,
                         "-sources",
                         "test_data=csv",
-                        "-source-readonly",
                         "-source-filename",
                         test_file,
                     ]
@@ -339,7 +335,6 @@ async def test_run(self):
                     *features,
                     "-sources",
                     "predict_data=csv",
-                    "-source-readonly",
                     "-source-filename",
                     predict_file,
                 )