def write_comprehend_sentiment()

in python/ts-to-word.py [0:0]

• 57 lines of code
• 3 McCabe index (conditional complexity)

def write_comprehend_sentiment(document, speech_segments, temp_files):
    """
    Writes out tables for per-period, per-speaker sentiment from the analytics mode, as well as
    the overall sentiment for a speaker

    :param document: Docx document to add the sentiment graph to
    :param speech_segments: Process transcript text holding turn-by-turn sentiment
    :param temp_files: List of temp files to be deleted later
    :return:
    """
    # Initialise our base structures
    speaker0labels = ['ch_0', 'spk_0']
    speaker1labels = ['ch_1', 'spk_1']
    speaker0timestamps = []
    speaker0data = []
    speaker1timestamps = []
    speaker1data = []

    # Start with some spacing and a new sub-header
    document.add_paragraph()
    write_custom_text_header(document, "Amazon Comprehend Sentiment")
    # Now step through and process each speech segment's sentiment
    for segment in speech_segments:
        if segment.segmentIsPositive or segment.segmentIsNegative:
            # Only interested in actual sentiment entries
            score = segment.segmentSentimentScore
            timestamp = segment.segmentStartTime

            # Positive re-calculation
            if segment.segmentIsPositive:
                score = 2 * ((1 - (1 - score) / (1 - MIN_SENTIMENT_POSITIVE)) * 0.5)
            # Negative re-calculation
            else:
                score = 2 * ((1 - score) / (1 - MIN_SENTIMENT_NEGATIVE) * 0.5 - 0.5)

            if segment.segmentSpeaker in speaker1labels:
                speaker1data.append(score)
                speaker1timestamps.append(timestamp)
            elif segment.segmentSpeaker in speaker0labels:
                speaker0data.append(score)
                speaker0timestamps.append(timestamp)

    # Spline fit needs at least 4 points for k=3, but 5 works better
    speaker1k = 3
    speaker0k = 3
    if len(speaker1data) < 5:
        speaker1k = 1
    if len(speaker0data) < 5:
        speaker0k = 1

    # Create Speaker-0 graph
    plt.figure(figsize=(8, 5))
    speaker0xnew = np.linspace(speaker0timestamps[0], speaker0timestamps[-1],
                               int((speaker0timestamps[-1] - speaker0timestamps[0]) + 1.0))
    speaker0spl = make_interp_spline(speaker0timestamps, speaker0data, k=speaker0k)
    speaker0powerSmooth = speaker0spl(speaker0xnew)
    plt.plot(speaker0timestamps, speaker0data, "ro")
    plt.plot(speaker0xnew, speaker0powerSmooth, "r", label="Speaker 1")

    # Create Speaker-1 graph
    speaker1xnew = np.linspace(speaker1timestamps[0], speaker1timestamps[-1],
                               int((speaker1timestamps[-1] - speaker1timestamps[0]) + 1.0))
    speaker1spl = make_interp_spline(speaker1timestamps, speaker1data, k=speaker1k)
    speaker1powerSmooth = speaker1spl(speaker1xnew)
    plt.plot(speaker1timestamps, speaker1data, "bo")
    plt.plot(speaker1xnew, speaker1powerSmooth, "b", label="Speaker 2")

    # Draw it out
    plt.title("Call Sentiment - Pos/Neg Only")
    plt.xlabel("Time (seconds)")
    plt.axis([0, max(speaker0timestamps[-1], speaker1timestamps[-1]), -1.5, 1.5])
    plt.legend()
    plt.axhline(y=0, color='k')
    plt.axvline(x=0, color='k')
    plt.grid(True)
    plt.xticks(np.arange(0, max(speaker0timestamps[-1], speaker1timestamps[-1]), 60))
    plt.yticks(np.arange(-1, 1.01, 0.25))

    # Write out the chart
    chart_file_name = "./" + "sentiment.png"
    plt.savefig(chart_file_name)
    temp_files.append(chart_file_name)
    plt.clf()
    document.add_picture(chart_file_name, width=Cm(14.64))
    document.paragraphs[-1].alignment = WD_ALIGN_PARAGRAPH.LEFT