Skip to main content

04. Spanish sentiment (positive range)

sentiment_range_spanish_alt_python(left_end=0.9, right_end=1.0), selects rows whose Spanish text scores ≥0.9 from the sentiment-analysis-spanish TensorFlow/Keras model. The model is small but loading and running it is several seconds, so this example is marked slow in addition to docker.

What you'll see​

Loaded 10 tweets.

Rows scoring in [0.9, 1.0]:
                              text   score
   excelente maravilloso fantástico  0.987
       muy bueno me encanta totalmente  0.974
   magnífico extraordinario lo mejor  0.962

(Exact scores will vary slightly between runs, the model isn't deterministic across TF versions.)

The code​

Rows 0 to 4 carry deliberately positive Spanish phrases so the [0.9, 1.0] window always returns a non-empty result; rows 5 to 9 are neutral fillers:

_ROWS = [
    ('2022-01-01T00:00:00', 'excelente maravilloso totalmente fantástico'),
    ('2022-01-01T01:00:00', 'muy bueno me encanta este trabajo magnífico'),
    ('2022-01-01T02:00:00', 'magnífico extraordinario lo mejor de todo'),
    ('2022-01-01T03:00:00', 'fantástico genial perfecto fenomenal increíble'),
    ('2022-01-01T04:00:00', 'increíble experiencia maravillosa felicidades'),
    ('2022-01-01T05:00:00', 'ciudad metro urbano social política cultura'),
    # ...4 more neutral rows...
]

Unlike the histogram tutorials, this analytic returns rows rather than a top-k:

from whistlerlib import Context

ctx = Context('processes', 'localhost', 8786)
ds = ctx.load_csv(
    filen=csv_path,
    meta={
        'column_mapping': {'date_column': 'Date', 'text_column': 'text'},
        'file_encoding': 'utf-8',
    },
    num_partitions=2,
)
print(f'Loaded {ds.tweet_count()} tweets.')
positive = ds.sentiment_range_spanish_alt_python(left_end=0.9, right_end=1.0)
print(positive.to_string(index=False))

sentiment_range_spanish_alt_python(left_end, right_end) cleans each row's text (URL, mention, hashtag, punctuation, stopword removal), feeds it to SentimentAnalysisSpanish().sentiment(...) (a TensorFlow/Keras model from the sentiment-analysis-spanish package, score in [0, 1]), keeps rows whose score falls in [left_end, right_end], and .compute()s the survivors to a pandas DataFrame. The model is lazily loaded on each worker the first time the partition closure runs; that's why this tutorial carries the slow marker in addition to docker.

The full file (including the tempfile setup and CLI shim) is at examples/04-sentiment-spanish/example.py.

Run it​

First run downloads the sentiment-analysis-spanish model and may take a couple of minutes; subsequent runs reuse the cached weights.

# From examples/04-sentiment-spanish/, bring up a local Dask cluster, run the example, tear it down.
docker compose -f ../../docker/docker-compose.yml up -d
python example.py
docker compose -f ../../docker/docker-compose.yml down