![14
1.3 <= 1.2
# Read text file
dataset = (tf.contrib.data.TextLineDataset(file_path)
# Skip header row
.skip(1)
# Transform each elem by applying decode_csv fn
.map(decode_csv))
# Randomize input
dataset = dataset.shuffle(buffer_size=256)
# Repeats dataset this # times
dataset = dataset.repeat(repeat_count)
# Batch size to use
dataset = dataset.batch(32)
iterator = dataset.make_one_shot_iterator()
batch_features, batch_labels = iterator.get_next()
# Load dataset
training_set =
tf.contrib.learn.datasets.base.load_csv_with_header(
filename=IRIS_TRAINING,
target_dtype=np.int,
features_dtype=np.float32)
feature_columns =
[tf.feature_column.numeric_column("x", shape=[4])]
# Define the training inputs
train_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": np.array(training_set.data)},
y=np.array(training_set.target),](https://image.slidesharecdn.com/ml2017-171011113410/85/Al-around-ML-2017-14-320.jpg)
![# разворачиваем processImage
gcloud beta functions deploy ocr-extract --stage-bucket [YOUR_STAGING_BUCKET_NAME] --trigger-bucket
[YOUR_IMAGE_BUCKET_NAME] --entry-point processImage
# разворачиваем translateText
gcloud beta functions deploy ocr-extract --stage-bucket [YOUR_STAGING_BUCKET_NAME] --trigger-bucket
[YOUR_IMAGE_BUCKET_NAME] --entry-point processImage
# разворачиваем saveResult
gcloud beta functions deploy ocr-save --stage-bucket [YOUR_TEXT_BUCKET_NAME] --trigger-topic
[YOUR_RESULT_TOPIC_NAME] --entry-point saveResult
25
# создаем бакеты для данных
gsutil mb gs://[YOUR_STAGING_BUCKET_NAME]
gsutil mb gs://[YOUR_IMAGE_BUCKET_NAME]
gsutil mb gs://[YOUR_TEXT_BUCKET_NAME]
# локально создаем папку проекта
mkdir ~/gcf_ocr
cd ~/gcf_ocr
# портируем index.js и package.json с Git
# создаем новый файл
config.json
{
"RESULT_TOPIC": "[YOUR_RESULT_TOPIC_NAME]",
"RESULT_BUCKET": "[YOUR_TEXT_BUCKET_NAME]",
"TRANSLATE_TOPIC":
"[YOUR_TRANSLATE_TOPIC_NAME]",
"TRANSLATE": true,
"TO_LANG": ["en", "fr", "es", "ja", "ru"]
}
# загружаем картинку
gsutil cp [PATH_TO_IMAGE] gs://[YOUR_IMAGE_BUCKET_NAME]
# проверяем
gcloud beta functions logs read --limit 100
LEVEL NAME EXECUTION_ID TIME_UTC LOG
D ocr-extract 155750695892074 2017-09-25 18:51:28.260 Function execution started
I ocr-extract 155750695892074 2017-09-25 18:51:28.649 Looking for text in image bender.png
I ocr-extract 155750695892074 2017-09-25 18:51:31.148 Extracted text from image (9 chars)
I ocr-extract 155750695892074 2017-09-25 18:51:31.722 Detected language "en" for bender.png
I ocr-extract 155750695892074 2017-09-25 18:51:32.944 File bender.png processed.
D ocr-extract 155750695892074 2017-09-25 18:51:32.950 Function execution took 4690 ms, finished with status: 'ok'
D ocr-translate 113818010410062 2017-09-25 18:51:34.044 Function execution started
D ocr-save 113818102696070 2017-09-25 18:51:34.276 Function execution started
I ocr-translate 113818010410062 2017-09-25 18:51:34.456 Translating text into ru
I ocr-save 113818102696070 2017-09-25 18:51:34.680 Received request to save file bender.png
I ocr-save 113818102696070 2017-09-25 18:51:34.769 Saving result to bender.png_to_en.txt in bucket txtbuck
I ocr-translate 113818010410062 2017-09-25 18:51:35.921 Text translated to ru
D ocr-translate 113818010410062 2017-09-25 18:51:35.928 Function execution took 1885 ms, finished with status: 'ok'
...
I ocr-save 113817650764455 2017-09-25 18:51:48.919 File saved.
D ocr-save 113817650764455 2017-09-25 18:51:48.967 Function execution took 7642 ms, finished with status: 'ok'](https://image.slidesharecdn.com/ml2017-171011113410/85/Al-around-ML-2017-25-320.jpg)
Al around ML 2017, Оксана Савенко, студентка НТУ имени Каразина
- 1. All around Machine Learning Tensorflow, ML APIs, Google Assistant Oksana Savenko student of Karazin Kharkiv National University ML is a world of capabilities
- 2. What’s it about? ● Recommendation systems ● Predictions ● Recognition abilities ● Diagnostics ● Anomaly detection
- 3. ML in Google
- 4. ML in Google
- 5. ++
- 7. Neural Network Classifier Iris Setosa Iris Versicolor Iris Virginica
- 8. Code 8
- 11. Deep Playground
- 13. 13
- 14. 14 1.3 <= 1.2 # Read text file dataset = (tf.contrib.data.TextLineDataset(file_path) # Skip header row .skip(1) # Transform each elem by applying decode_csv fn .map(decode_csv)) # Randomize input dataset = dataset.shuffle(buffer_size=256) # Repeats dataset this # times dataset = dataset.repeat(repeat_count) # Batch size to use dataset = dataset.batch(32) iterator = dataset.make_one_shot_iterator() batch_features, batch_labels = iterator.get_next() # Load dataset training_set = tf.contrib.learn.datasets.base.load_csv_with_header( filename=IRIS_TRAINING, target_dtype=np.int, features_dtype=np.float32) feature_columns = [tf.feature_column.numeric_column("x", shape=[4])] # Define the training inputs train_input_fn = tf.estimator.inputs.numpy_input_fn( x={"x": np.array(training_set.data)}, y=np.array(training_set.target),
- 15. Cloud ML Engine
- 16. Google Cloud Machine Learning Use Cases
- 17. MNIST 17
- 18. MNIST 18
- 19. TensorBoard 19
- 20. ML APIs
- 24. Double Kill Optical Character Recognition (OCP) Google Cloud Functions Google Cloud Pub/Sub Google Cloud Storage Google Cloud Translation API Google Cloud Vision API
- 25. # разворачиваем processImage gcloud beta functions deploy ocr-extract --stage-bucket [YOUR_STAGING_BUCKET_NAME] --trigger-bucket [YOUR_IMAGE_BUCKET_NAME] --entry-point processImage # разворачиваем translateText gcloud beta functions deploy ocr-extract --stage-bucket [YOUR_STAGING_BUCKET_NAME] --trigger-bucket [YOUR_IMAGE_BUCKET_NAME] --entry-point processImage # разворачиваем saveResult gcloud beta functions deploy ocr-save --stage-bucket [YOUR_TEXT_BUCKET_NAME] --trigger-topic [YOUR_RESULT_TOPIC_NAME] --entry-point saveResult 25 # создаем бакеты для данных gsutil mb gs://[YOUR_STAGING_BUCKET_NAME] gsutil mb gs://[YOUR_IMAGE_BUCKET_NAME] gsutil mb gs://[YOUR_TEXT_BUCKET_NAME] # локально создаем папку проекта mkdir ~/gcf_ocr cd ~/gcf_ocr # портируем index.js и package.json с Git # создаем новый файл config.json { "RESULT_TOPIC": "[YOUR_RESULT_TOPIC_NAME]", "RESULT_BUCKET": "[YOUR_TEXT_BUCKET_NAME]", "TRANSLATE_TOPIC": "[YOUR_TRANSLATE_TOPIC_NAME]", "TRANSLATE": true, "TO_LANG": ["en", "fr", "es", "ja", "ru"] } # загружаем картинку gsutil cp [PATH_TO_IMAGE] gs://[YOUR_IMAGE_BUCKET_NAME] # проверяем gcloud beta functions logs read --limit 100 LEVEL NAME EXECUTION_ID TIME_UTC LOG D ocr-extract 155750695892074 2017-09-25 18:51:28.260 Function execution started I ocr-extract 155750695892074 2017-09-25 18:51:28.649 Looking for text in image bender.png I ocr-extract 155750695892074 2017-09-25 18:51:31.148 Extracted text from image (9 chars) I ocr-extract 155750695892074 2017-09-25 18:51:31.722 Detected language "en" for bender.png I ocr-extract 155750695892074 2017-09-25 18:51:32.944 File bender.png processed. D ocr-extract 155750695892074 2017-09-25 18:51:32.950 Function execution took 4690 ms, finished with status: 'ok' D ocr-translate 113818010410062 2017-09-25 18:51:34.044 Function execution started D ocr-save 113818102696070 2017-09-25 18:51:34.276 Function execution started I ocr-translate 113818010410062 2017-09-25 18:51:34.456 Translating text into ru I ocr-save 113818102696070 2017-09-25 18:51:34.680 Received request to save file bender.png I ocr-save 113818102696070 2017-09-25 18:51:34.769 Saving result to bender.png_to_en.txt in bucket txtbuck I ocr-translate 113818010410062 2017-09-25 18:51:35.921 Text translated to ru D ocr-translate 113818010410062 2017-09-25 18:51:35.928 Function execution took 1885 ms, finished with status: 'ok' ... I ocr-save 113817650764455 2017-09-25 18:51:48.919 File saved. D ocr-save 113817650764455 2017-09-25 18:51:48.967 Function execution took 7642 ms, finished with status: 'ok'
- 26. 26 APPLAUDISSEMENTS APPLAUSE 拍手 APLAUSOS АПЛОДИСМЕНТЫ
- 27. Video Intelligence API BETA Features ● label detection ● shot change detection ● regionalization
- 28. Speech API
- 31. Jobs API BETA ● check better job site engagement ● check higher apply rates ● check increased candidate conversion ● check freed up product and engineering resources
- 33. 2 Ways to integrate Google Assistant library Hands-free activation - activate your device by saying Hey Google or Ok Google, just like with Google Home! Audio capture and playback Conversation state management Timer and alarm management Google Assistant gRPС API Idiomatic client libraries in 10 languages Highly efficient on wire and with a simple service definition framework Bi-directional streaming with http/2 based transport Pluggable auth, tracing, load balancing and health checking Google Assistant gRPC API
- 34. Conversation Design Create a persona Pick up the voice
- 35. Tricks ● Unlock Your Phone with Google Assistant ● Get Google Assistant to Remember Things ● Take a Selfie ● See Your Photos ● Post to Facebook by Voice ● Translate on the Go https://ifttt.com/join 35
- 36. 36 Google News наушники Pixel Buds
- 37. 37 Google News Teachable Machine https://teachablemachine.withgoogle.com/
- 38. 38