Python statistics.stdev方法代码示例

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def evaluate_and_update_max_score(self, t, episodes):
        eval_stats = eval_performance(
            self.env, self.agent, self.n_steps, self.n_episodes,
            max_episode_len=self.max_episode_len,
            logger=self.logger)
        elapsed = time.time() - self.start_time
        custom_values = tuple(tup[1] for tup in self.agent.get_statistics())
        mean = eval_stats['mean']
        values = (t,
                  episodes,
                  elapsed,
                  mean,
                  eval_stats['median'],
                  eval_stats['stdev'],
                  eval_stats['max'],
                  eval_stats['min']) + custom_values
        record_stats(self.outdir, values)
        if mean > self.max_score:
            self.logger.info('The best score is updated %s -> %s',
                             self.max_score, mean)
            self.max_score = mean
            if self.save_best_so_far_agent:
                save_agent(self.agent, "best", self.outdir, self.logger)
        return mean 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def calc_disagreement(evaluations):
    """Return the disagreement level for evaluations, or None if no evaluations.

    Calculated as the max disagreement of (1) N/A and non-N/A responses and (2) non-N/A evaluations
    :param evaluations: an iterable of Eval
    """
    if evaluations:
        na_it, rated_it = partition(lambda x: x is not Eval.not_applicable, evaluations)
        na_votes = list(na_it)
        rated_votes = list(rated_it)

        # Here we use the sample standard deviation because we consider the evaluations are a sample of all the
        # evaluations that could be given.
        # Not clear the best way to make the N/A disagreement comparable to the evaluation disagreement calculation
        na_disagreement = (
            statistics.stdev(([0] * len(na_votes)) + ([1] * len(rated_votes)))
            if len(na_votes) + len(rated_votes) > 1
            else 0.0)
        rated_disagreement = (
            statistics.stdev([v.value for v in rated_votes])
            if len(rated_votes) > 1
            else 0.0)
        return max(na_disagreement, rated_disagreement)
    else:
        return None 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def _random(self, env, min_mean, max_mean, min_stdev, max_stdev,
                test_count=3, sample_count=300):
        mmin, smin, mmax, smax = 100, 100, 0, 0
        for i in range(test_count):
            values = [env(0) for i in range(sample_count)]
            mean, stdev = statistics.mean(values), statistics.stdev(values)
            mmax = max(mmax, mean)
            mmin = min(mmin, mean)
            smax = max(smax, stdev)
            smin = min(smin, stdev)

        self.assertGreater(mmin, min_mean)
        self.assertLess(mmax, max_mean)
        self.assertGreater(smin, min_stdev)
        self.assertLess(smax, max_stdev)

        return mmin, mmax, smin, smax 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def eval_performance(process_idx, make_env, model, phi, n_runs):
    assert n_runs > 1, 'Computing stdev requires at least two runs'
    scores = []
    for i in range(n_runs):
        model.reset_state()
        env = make_env(process_idx, test=True)
        obs = env.reset()
        done = False
        test_r = 0
        while not done:
            s = chainer.Variable(np.expand_dims(phi(obs), 0))
            pout, _ = model.pi_and_v(s)
            a = pout.action_indices[0]
            obs, r, done, info = env.step(a)
            test_r += r
        scores.append(test_r)
        print('test_{}:'.format(i), test_r)
    mean = statistics.mean(scores)
    median = statistics.median(scores)
    stdev = statistics.stdev(scores)
    return mean, median, stdev 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def eval_performance(rom, p_func, n_runs):
    assert n_runs > 1, 'Computing stdev requires at least two runs'
    scores = []
    for i in range(n_runs):
        env = ale.ALE(rom, treat_life_lost_as_terminal=False)
        test_r = 0
        while not env.is_terminal:
            s = chainer.Variable(np.expand_dims(dqn_phi(env.state), 0))
            pout = p_func(s)
            a = pout.action_indices[0]
            test_r += env.receive_action(a)
        scores.append(test_r)
        print('test_{}:'.format(i), test_r)
    mean = statistics.mean(scores)
    median = statistics.median(scores)
    stdev = statistics.stdev(scores)
    return mean, median, stdev 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def to_dict(self):
        return {
            "LB_min": round(self.latency.min, 2),
            "LB_avg": round(self.latency.avg, 2),
            "LB_max": round(self.latency.max, 2),
            "L_stdev": round(self.latency.stdev, 2),
            "L_stdev[%]": round(self.latency.stdev / self.latency.avg * 100, 2),
            # ---
            "TB_min": round(self.throughput.min, 2),
            "TB_avg": round(self.throughput.avg, 2),
            "TB_max": round(self.throughput.max, 2),
            "T_stdev": round(self.throughput.stdev, 2),
            "T_stdev[%]": round(self.throughput.stdev / self.throughput.avg * 100, 2),
            # ---
            "B_count": self.count,
            "app": self.name
        } 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def get_partitions_info_str(j):
    partitions = j['components']['partition_counts']['counts']
    partitions_info = {
                          'Partitions': len(partitions),
                          'Rows': sum(partitions),
                          'Empty partitions': len([p for p in partitions if p == 0])
                      }
    if partitions_info['Partitions'] > 1:
        partitions_info.update({
            'Min(rows/partition)': min(partitions),
            'Max(rows/partition)': max(partitions),
            'Median(rows/partition)': median(partitions),
            'Mean(rows/partition)': int(mean(partitions)),
            'StdDev(rows/partition)': int(stdev(partitions))
        })


    return "\n{}".format(IDENT).join(['{}: {}'.format(k, v) for k, v in partitions_info.items()]) 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def status(self):
        import collections, statistics
        from datetime import datetime
        return {
            "starttime": datetime.utcfromtimestamp(self.stats["start_time"]).strftime('%Y-%m-%d %H:%M:%S'),
            "pass": self.numPass(),
            "fail": self.numFails(),
            "failures": self.failures,
            "speed": self.testsPerSecond(),
            "mean": statistics.mean(self.traceLengths) if self.traceLengths else "NA",
            "stdev": statistics.stdev(self.traceLengths) if len(self.traceLengths) > 2 else "NA",
            "numZero": self.traceLengths.count(0) if self.traceLengths else "NA",
            "max": max(self.traceLengths) if self.traceLengths else "NA",
            "maxDepth": max(self.traceDepths) if self.traceDepths else "NA",
            "numConst": statistics.mean(self.traceConstantinopleOps) if self.traceConstantinopleOps else "NA",
            "activeSockets": self.stats["num_active_sockets"],
            "activeTests": self.stats["num_active_tests"],
        } 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def evaluate_and_update_max_score(self, t, episodes):
        eval_stats = eval_performance(
            self.env, self.agent, self.n_runs,
            max_episode_len=self.max_episode_len, explorer=self.explorer,
            logger=self.logger)
        elapsed = time.time() - self.start_time
        custom_values = tuple(tup[1] for tup in self.agent.get_statistics())
        mean = eval_stats['mean']
        values = (t,
                  episodes,
                  elapsed,
                  mean,
                  eval_stats['median'],
                  eval_stats['stdev'],
                  eval_stats['max'],
                  eval_stats['min']) + custom_values
        record_stats(self.outdir, values)
        if mean > self.max_score:
            self.logger.info('The best score is updated %s -> %s',
                             self.max_score, mean)
            self.max_score = mean
            if self.save_best_so_far_agent:
                save_agent(self.agent, t, self.outdir, self.logger)
        return mean 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def within_stdev_percent(values, x_stdev, percent_threshold, min_values=100):
    '''Return True if percent_threshold of values are within x_stdev of the mean.'''
    if len(values) < min_values:
        return True

    mean = statistics.mean(values)
    stdev = statistics.stdev(values)
    found = []
    for v in values:
        diff = abs(mean - v)
        if diff <= (stdev * x_stdev):
            found.append(v)
    percent_found = len(found) / len(values)
    result = percent_found > percent_threshold
    log.debug(f"Within {x_stdev} sigma check was {result}. {percent_found:.2f}%/{percent_threshold:.2f}% within stdev*{x_stdev}. "
              f"Mean: {mean:.2f}. Stdev: {stdev:.2f}. Acceptable range was: {mean - stdev * x_stdev:.2f} - {mean + stdev * x_stdev:.2f}")
    return result 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def test_ci95(self):
        for length in [2, 3, 5, 10, 100, 1000]:
            numbers = [random.random() for _ in range(length)]
            ci = plot.ci95(numbers)
            mu = mean(numbers)
            std = stdev(numbers, xbar=mu)
            lower = mu - 2.0 * std / math.sqrt(length)
            upper = mu + 2.0 * std / math.sqrt(length)
            self.assertTrue(ci[0] - lower <= 1e-6)
            self.assertTrue(ci[1] - upper <= 1e-6)

            # Test the documentation example
            smoothed = []
            for replay in range(10):
                rewards = [random.random() for _ in range(100)]
                y_smoothed = plot.smooth(rewards)
                smoothed.append(y_smoothed)
            means = [mean(r) for r in zip(*smoothed)]
            confidences = [plot.ci95(r) for r in zip(*smoothed)]
            lower_bounds = [conf[0] for conf in confidences]
            upper_bounds = [conf[1] for conf in confidences]
            for lb, ub, m in zip(lower_bounds, upper_bounds, means):
                self.assertTrue(lb <= m)
                self.assertTrue(ub >= m) 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def cohens(dataset, predictions, combined_data: CombinedData):
    xs = combined_data.get_explanatory_variables()
    ys = combined_data.get_explained_variables()
    x = xs[0]
    y = ys[0]
    cat = [k for k,v in x.metadata[categories].items()]
    data = []

    pred = None
    if predictions:
        pred = predictions[0][0]

    lhs = None
    rhs = None
    for c in cat:
        cat_data = dataset.select(y.metadata[name], where=[f"{x.metadata[name]} == '{c}'"])
        if c == pred.lhs.value:
            lhs = cat_data
        if c == pred.rhs.value:
            rhs = cat_data
        data.append(cat_data)

    cohens_d = (mean(lhs) - mean(rhs)) / (sqrt((stdev(lhs) ** 2 + stdev(rhs) ** 2) / 2))
    return cohens_d 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def _count_resource(self, attr_name, aggr_function=None) -> Tuple[float, float]:
        """
        Calculate resources from exec reports.

        :param attr_name: name of the attribute of execreport to count resource.
        :param aggr_function:  function to process value of execreport.

        :return: (mean_value, standart_deviation)
        """
        if not aggr_function:
            aggr_function = lambda x: x  # noqa: E731

        values = [aggr_function(getattr(i, attr_name)) for i in self.exec_reports]
        mean_value = mean(values)
        std_dev = stdev(values) if len(values) > 1 else 0

        return (mean_value, std_dev) 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def evaluate():
    model.eval() # Turn on the evaluation mode
    with torch.no_grad():
        # evaluating
        node_embeddings = model.ss.weight
        graph_embeddings = torch.spmm(graph_pool, node_embeddings).data.cpu().numpy()
        acc_10folds = []
        for fold_idx in range(10):
            train_idx, test_idx = separate_data_idx(graphs, fold_idx)
            train_graph_embeddings = graph_embeddings[train_idx]
            test_graph_embeddings = graph_embeddings[test_idx]
            train_labels = graph_labels[train_idx]
            test_labels = graph_labels[test_idx]

            cls = LogisticRegression(solver="liblinear", tol=0.001)
            cls.fit(train_graph_embeddings, train_labels)
            ACC = cls.score(test_graph_embeddings, test_labels)
            acc_10folds.append(ACC)
            print('epoch ', epoch, ' fold ', fold_idx, ' acc ', ACC)

        mean_10folds = statistics.mean(acc_10folds)
        std_10folds = statistics.stdev(acc_10folds)
        # print('epoch ', epoch, ' mean: ', str(mean_10folds), ' std: ', str(std_10folds))

    return mean_10folds, std_10folds 

# 需要导入模块: import statistics [as 别名]
# 或者: from statistics import stdev [as 别名]
def eval_performance(env, agent, n_steps, n_episodes, max_episode_len=None,
                     logger=None):
    """Run multiple evaluation episodes and return statistics.

    Args:
        env (Environment): Environment used for evaluation
        agent (Agent): Agent to evaluate.
        n_steps (int): Number of timesteps to evaluate for.
        n_episodes (int): Number of evaluation episodes.
        max_episode_len (int or None): If specified, episodes longer than this
            value will be truncated.
        logger (Logger or None): If specified, the given Logger object will be
            used for logging results. If not specified, the default logger of
            this module will be used.
    Returns:
        Dict of statistics.
    """

    assert (n_steps is None) != (n_episodes is None)

    if isinstance(env, chainerrl.env.VectorEnv):
        scores = batch_run_evaluation_episodes(
            env, agent, n_steps, n_episodes,
            max_episode_len=max_episode_len,
            logger=logger)
    else:
        scores = run_evaluation_episodes(
            env, agent, n_steps, n_episodes,
            max_episode_len=max_episode_len,
            logger=logger)
    stats = dict(
        episodes=len(scores),
        mean=statistics.mean(scores),
        median=statistics.median(scores),
        stdev=statistics.stdev(scores) if len(scores) >= 2 else 0.0,
        max=np.max(scores),
        min=np.min(scores))
    return stats