Simple labeling and analysis code for my cs294-10 Visualization project, Fall 2013.

assign_nuggets.py

import sys
import csv
import os

import numpy as N
import vincent
import simplejson as json

# voiceboxmongo provides a connection to my database,
# which stores data from Mechanical Turk
import voiceboxmongo


def load_statements(vis, strategy=None):
    with voiceboxmongo.VoiceboxMongo() as client:
        db = client.da_comparisons
        query = {"vis": vis}
        if strategy is not None:
            query["strategy"] = strategy
        results = db.visstratresults.find(query)
        return [res for res in results]


def assign_nuggets(vis):
    statements = [s["response"][0] for s in load_statements(vis, strategy="E")]

    with open("data/vis/%s/nug_scores.csv" % vis, 'rb') as f:
        reader = csv.reader(f)
        for row in reader:
            header = row
            break

    header.append('WRONG')

    results = N.zeros((len(statements), len(header)))

    for si, s in enumerate(statements):
        os.system('clear')
        for i, h in enumerate(header):
            print "%d\t%s" % (i, h)

        print
        print "%d. %s" % (si, s)

        nugs = raw_input("Contained nuggets (space separated): ")
        nugs = nugs.split(' ')
        try:
            nugs = N.array([int(s) for s in nugs])
            results[si][nugs] = 1
            print
            print results[si]
        except:
            pass

    N.savetxt("data/vis/%s/nug_assignments.csv" % vis, results, delimiter=',')

    build_nugget_result(vis, statements, wrong_col=True)


def build_nugget_result(vis, statements, wrong_col=False):

    nug_weights = N.loadtxt("data/vis/%s/nug_scores.csv" % vis, delimiter=',', skiprows=1)
    nugs = N.loadtxt("data/vis/%s/nug_assignments.csv" % vis, delimiter=',')

    # remove the "wrong" column for now
    if wrong_col:
        nugs = nugs[:, :-1]

    print vis, "response scores"

    scores = N.nan_to_num(N.dot(nugs, nug_weights) / N.sum(nugs, axis=1))
    hist = N.histogram(scores, bins=10, range=(0,1))
    print "median score", N.median(scores)
    df = {"v": list(hist[0]), 
          "i": [float("%02f" % s) for s in list((hist[1][1:] + hist[1][:-1])/2.0)] }
    score_histogram = vincent.Bar(df, iter_idx="i", key_on="v", height=200, width=400)
    # score_histogram.display()

    nug_distribution = -N.sort(-N.sum(nugs, axis=0))
    dist_bar = vincent.Bar(list(nug_distribution), height=200, width=400)
    dist_bar.scales[0].range_max = len(nug_distribution) * 25
    # dist_bar.display()

    info = {}

    with open("data/vis/%s/nug_scores.csv" % vis, 'rb') as f:
        reader = csv.reader(f)
        for row in reader:
            nuggets = row
            break

    info["statements"] = []
    for i in range(len(statements)):
        nug_parts = []
        for j in range(len(nuggets)):
            if nugs[i, j] != 0:
                nug_parts.append(j)
        info["statements"].append({
            "statement": statements[i],
            "nuggets": nug_parts,
            "score": scores[i]
        })
    info["nuggets"] = []

    for i in range(len(nuggets)):
        cluster = []
        for j in range(len(statements)):
            if nugs[j, i] != 0:
                cluster.append(j)

        info["nuggets"].append({
            "score": nug_weights[i],
            "nugget": nuggets[i],
            "statements": cluster
        })
    info["score_histogram"] = {
        "bin": df["i"],
        "count": df["v"],
        "spec": score_histogram.grammar()
    }
        
    json.dump(info, open("data/vis/%s/result.csv" % vis, 'w'))


if __name__ == '__main__':
    if len(sys.argv) > 2:
        if sys.argv[2] == 'skip':
            wrong_col = False
            if len(sys.argv) > 3:
                if sys.argv[3] == 'w':
                    wrong_col = True

            vis = sys.argv[1]
            statements = [s["response"][0] for s in load_statements(vis, strategy="E")]
            build_nugget_result(vis, statements, wrong_col=wrong_col)
    else:
        assign_nuggets(sys.argv[1])