Aller au contenu

Messages recommandés

Il y a 2 heures, Lancelot a dit :

Bah puisque tu as tout bien compris dis moi, il était où le goalpost au début et il est où maintenant ?

 

Ici par exemple:

 

Il y a 5 heures, Lancelot a dit :

Pris la main non, rouler une pelle ça commence à pouvoir faire monter la sauce.

 

Et si tu n'es pas d'accord:

 

Il y a 5 heures, Lancelot a dit :

Relis la conversation au calme dans quelques jours, je ne vois pas quoi te dire d'autre :jesaispo:

 

Lien vers le commentaire
35 minutes ago, Hayek's plosive said:

Ici par exemple:

Oui et ? Tu ne réponds pas à ma question.

 

35 minutes ago, Hayek's plosive said:

Et si tu n'es pas d'accord:

Avant d'être d'accord ou pas il faudrait que je comprenne ce que tu essaies de communiquer, là j'avoue que ce n'est pas clair.

Lien vers le commentaire
8 hours ago, poney said:

Est ce que se prendre par la main c'est tromper ?

Tu veux dire en se fistant ?

(Oui je sais j'abaisse le niveau de finesse).

  • Haha 1
  • Nay 1
Lien vers le commentaire

Tout passera au moule féministe !

 Déjà elle commence par dire "j'ai essayé d'organiser un peu ma pensée pour cette vidéo" -> elle fait un simple plan thématique en 3 parties (c'est dire le niveau habituel)

Donc en quoi l'architecture serait sexiste :

  1. L’Enseignement est sexiste : car 59% des étudiants en architecture sont des femmes !
  2. Le marché du travail est sexiste car les architectes femmes sont en moyenne 2x moins payée (mais elle dit elle même que les femmes sont plus jeunes et davantage salariées ou fonctionnaires tandis que les hommes sont plus souvent à leur compte...serait-ce un début d'explication?)
  3. La représentation de l'architecte dans la culture est sexiste car la femme est vue comme une muse ou pire une cliente ! (et aussi les hommes s'approprient le travail de leur épouse lors des remises de prix ? quel rapport avec la culture + c'est faux puisque les époux mentionnent bel et bien leur épouse dans leur remerciements?)
Lien vers le commentaire
25 minutes ago, Bisounours said:

C'est encourageant mais vous pouvez mieux faire.

J'avais "prendre un enfant par la main" mais je me suis dit que j'allais attendre deux trois surenchères avant.

Lien vers le commentaire
il y a 23 minutes, DeadBot a dit :

Un algorithme de prédiction du sexe en fonction de réponses à quelques questions :

Gender Continuum Test | Compare Your Personality To Others Of The Same Gender | ClearerThinking.org 

 

Attention, c'est assez long pour parvenir aux réponses si on veut tout lire. Certains points sur les différences hommes/femmes sont intéressants. 

"Il y a plein de petites différences pas très importantes entre les sexes", "c'est presque impossible d'inférer le sexe de quelqu'un d'après ses réponses au test", moui moui moui... Il n'empêche, même si mon anima est très forte pour un homme (je suis plus proche de la moyenne féminine que de la moyenne masculine pour 7 ou 8 items sur 18), la prédiction était correcte, et surtout le niveau de certitude atteignait 91 %, rien que ça. Donc je reste très sceptique sur le "impossible en général d'inférer le sexe d'après les réponses au questionnaire" : c'est certainement impossible d'après une ou deux réponses, mais d'après les 36 réponses on finit quand même par converger assez efficacement.

 

Je demande l'arbitrage de @Lancelot, évidemment.

Lien vers le commentaire
il y a 3 minutes, Jensen a dit :

Le plus gros biais de ce truc, c'est que c'est basé sur la façon dont les gens se décrivent eux-mêmes.

Oui, et même dans le cas où ils ne mentent pas, c'est aussi basé sur la manière dont ils comprennent les mots.

Lien vers le commentaire

J'aime bien leurs explications sur les stats.

Quote

The biggest correlation we found between gender and any one personality trait was r=0.42, which means that gender explains 18% of the variability in that trait. Another way to put this is... if you knew a random person's score on that one personality trait, and you tried to guess that person's gender using just that score, you'd mispredict that person's gender 31% of the time! So while men have higher scores on that trait, on average, than women do, there are still plenty of women who have higher scores than the majority of men.

Women are actually closer to the male average in this trait 37% of the time (i.e. they are more typically male than typically female in this trait). So that means that if there is a room of 100 women, about 37 of them will be closer to the average male score than they are to the average female score! And similarly, we found that men are closer to the female average than the male average 28% of the time.

And remember, that was the personality trait that we found to be most predictive of gender. All the other traits that had an association with gender had a substantially weaker relationship. For instance, gender had a correlation of r=0.30 with the second most predictive trait (so it is about 28% weaker as a predictor of gender than the strongest trait).

Despite finding no large differences, we did discover many small differences between men and women in personality! More specifically, we uncovered 18 different personality traits for which the average scores for men and women in the U.S. differed enough that we could demonstrate the effect reliably and repeatedly in our studies.

In other words, men and women seem to differ (on average) in self-reported personality in no large ways, but in many small ways!

While 18 may sound like a lot of differences, remember that we examined over 600 different personality questions looking for differences. On the vast majority of questions, we were not able to detect any differences in the answers of men and women.

For 9 of these 18 traits men had a higher average score than women, and for the other 9 women had a higher average score than men. So are "men like Mars" and "women like Venus"?

You may be surprised to find out that across all these personality traits, women were closer to the female average than the male average only 61% of the time, and men were closer to the male average than the female average only 57% of the time. In other words, it's extremely common for men to have quite a number of these "more female" traits, and for women to have quite a number of these "more male" traits. Almost all of us are a mix of both! In fact, only about 1% of males and 1% of females had almost entirely "more male" or almost entirely "more female" personality traits! So being a mix is by far the most common result. That means that, when it comes to how many of our personality traits are "more male" like vs. "more female" like, we're in a sense almost all somewhere within the "gender continuum", rather than totally at one end or the other.

Another interesting finding is that the list of traits that did not show meaningful gender differences is even longer than the list of traits that did show a difference!

[...]

You'll notice that while many men and women have scores in the range of 1.5 to 2.0, at the very highest end of being compassionate (a score of 3.0) there are about twice as many women. Even more dramatically, while very few people report themselves as being highly uncompassionate (with scores in the range of -1.5 to -3.0), nearly all of the people that do are men!

Imagine that you know the compassion scores of 100 people. The compassion scores alone would not allow you to guess with any reasonable level of accuracy who was male vs. female. This is because of the big overlap in the frequency of different compassion scores (especially in the range of 1.0 to 2.0) for men and women.

However, if you knew for a fact that a person had an extremely high compassion score (a score of 3), your guess that they are female would be about twice as likely to be true than if you guessed male. Likewise, if you knew that a person had a very low compassion score (e.g. -1.5 or lower) you would almost certainly be right if you guessed they are male!

And yet... there is only a small difference on average between male and female compassion scores.

So that means if you randomly pick a male and a female in the U.S. chances are they won't differ that much in their compassion. But when you only consider the most extreme outliers (i.e. people who are extremely compassionate or who very much lack compassion) you get a much more gendered result.

[...]

To give an example: while on average women score only a little higher than men on traits like being peaceful, compassionate and forgiving, if we consider statistics on mass shootings in the U.S., 96% of them were perpetrated by men. Similarly, males are convicted of the vast majority of homicides in the U.S., representing 90% of the total number of offenders. These cases being some of the most widely discussed examples of extreme violence, extreme lack of compassion, and extreme anger, perhaps these horrible events promote the stereotype of males typically being violent, uncompassionate and angry (despite the average differences between men and women in these traits actually being small).

But how do we then explain the really striking gender difference in who perpetuates mass shootings and violent homicides? A partial explanation could be that the act of committing horrendous acts of violence is associated extremely low levels of being peaceful, compassionate and forgiving, and while the averages for these traits are really not that different for men and women, men are far more likely than women to have extremely low levels of all three of these traits at once! So when we consider mass shootings, we're considering the extreme tail end of personality, and we're stacking on top of each other multiple traits that show strong gender differences at the extremes.

[...]

Recall that we could not predict gender from compassion scores above chance levels unless we considered only individuals with either very high or very low scores. But what if we knew an individual's scores on all the 18 traits that show small, but significant group differences between men and women? Could we guess who was male and who was female more than 50% of the time?

We asked humans and a machine algorithm to predict gender using information about how much an individual agreed or disagreed with 36 personality items, 2 for each of the 18 personality traits. Based on information such as "This individual... Strongly agrees with I worry a lot, Disagrees with I laugh aloud, Is Neutral on I love to solve complex problems, etc", the algorithm guessed gender correctly 78% of the time and humans guessed correctly 58% of the time. The algorithm was far better at making these predictions, but not perfect. Human accuracy was just above chance.

When we simplified the task for humans by showing only an individual's overall trait agreement (e.g., This individual... Strongly agrees with At-Ease, Disagrees with Warmth, Is Neutral on Complexity-Seeking) accuracy increased to just 60%. So if you knew a person's personality, but not their gender, you'd likely be very bad at predicting what gender they are!

Why can't humans use personality information to make accurate guesses about gender? We theorized that perhaps humans try to rely on gender stereotypes to make their guesses. As we've seen, most men and women score in the mid ranges on all 18 personality traits. The group differences are mostly due to extreme scores among men on the so-called "male traits" and extreme scores among women on the so-called "female traits". If we use stereotypes to make our guesses about gender, we will be wrong most of the time because most men and women are not extreme scorers on even a few of the 18 personality traits we studied.

We tested this theory by finding the handful of cases where men and women agreed at the extremes with the personality traits as dictated by social stereotypes. We asked humans to guess the gender of these individuals and found that accuracy increased to 72%! We knew that people are well aware of personality stereotypes from our visualization task (people are more likely to visualize an extremely unselfish, forgiving, compassionate person as a woman, and an extremely selfish, angry, uncompassionate person as a male). Apparently, we use these unreliable stereotypes to make predictions about gender and the result is that our predictions are only slightly better than chance.

La dernière partie est intéressante si on compare à ce que dit Murray (en gros ok la différence sur chaque trait est peu prédictive mais si on prend le pattern formé par tous les traits ça peut le devenir très significativement). Your mileage may vary sur le 72% d'accuracy et à quel point c'est significatif, j'imagine (leur modèle est une régression logistique linéaire, ils disent qu'ils ont testé des modèles non linéaires qui ne marchent pas mieux, après faudrait donner ça à un machine learning wizard pour voir).

 

En tous cas eux concluent "on rejette à la fois ceux qui disent que les hommes et les femmes sont complètement différents et ceux qui disent qu'ils n'y a aucune différence". Ce qui est une conclusion prudente, disons.

Quote

The idea that men and women are completely different doesn't hold up in our research. Remember, people's answers to many hundreds of personality questions showed differences on only 18 traits and the differences were all small.

In reality, almost everyone has some personality traits that are stereotypically "masculine" and some personality traits that are stereotypically "feminine". Claims that "men are like this" and "women are like that" are simply not supported by the data.

[...]

The idea that men and women are exactly the same also doesn't hold up in our personality research. Remember, we found 18 small, but statistically significant gender differences in personality. And we confirmed each of these differences in at least two studies, just to make sure they are real!

In addition, although a person's score on just one of the 18 personality traits isn't enough to predict gender with any significant accuracy, all 18 scores together provide enough information to predict gender correctly approximately 8 out of 10 times (i.e. with 78% accuracy in the case of our predictive model, which was trained using some of our data and then tested on the remainder of the data to produce an unbiased estimate).

 

Quote

You guessed 16 out of 18 correct, which means you were 88.89% accurate! For comparison, the average accuracy rate is 81%.

Fuck yeah! (hint: j'ai fait une erreur en inversant le genre entre "Self-Defending - unlikely to assign self-blame" et "Aesthetic - appreciating artistic beauty").

 

Quote

Based on your answers to the personality questions....

Our machine learning model predicts that you have a:

  • 99.32% chance of being Male
  • 0.68% chance of being Female

No surprise here.

De manière intéressante le seul trait "non conforme" que j'ai (en fait je suis entre les deux moyennes) est l'esthétique, ce qui peut expliquer en partie mon erreur plus haut.

 

J'aime bien leur présentation en termes de distribution de fréquences, on voit bien que c'est rarement régulier.

 

Au final ils font un travail de vulgarisation assez phénoménal, chapeau à eux.

  • Yea 2
Lien vers le commentaire
Il y a 9 heures, DeadBot a dit :

Un algorithme de prédiction du sexe en fonction de réponses à quelques questions :

Gender Continuum Test | Compare Your Personality To Others Of The Same Gender | ClearerThinking.org 

 

Attention, c'est assez long pour parvenir aux réponses si on veut tout lire. Certains points sur les différences hommes/femmes sont intéressants. 

Il avait reussi à prédire mon sexe avec 99% de certitude mais pas celui d'une amie (qui s'était vue attribuée à tort le sexe mâle)

Lien vers le commentaire

Le truc m'as vu à 55% nana 45% mec. Ce qui veut dire d'après leur propre explication que l'algo avait un très bas niveau de confiance dans son estimation et n'y est pas arrivé vu mes réponses. Bien que j'ai peu répondu de manière extrême dans le spectre de leur questions. (Je me demande si j'ai pas missclick un truc en mettant le plus haut score par accident mais bref).

Lien vers le commentaire
il y a 17 minutes, Alchimi a dit :

Le truc m'as vu à 55% nana 45% mec. Ce qui veut dire d'après leur propre explication que l'algo avait un très bas niveau de confiance dans son estimation et n'y est pas arrivé vu mes réponses

Ben non y'a pas d'info de précision https://fr.wikipedia.org/wiki/Exactitude_et_précision

Ptêtre que t'es vraiment au milieu selon leur échelle 

Lien vers le commentaire
il y a 2 minutes, ttoinou a dit :

Ben non y'a pas d'info de précision https://fr.wikipedia.org/wiki/Exactitude_et_précision

Ptêtre que t'es vraiment au milieu selon leur échelle 

Quelque part dans les explications ils disent que si l'algo donne un quelconque résultat avec une marge de 55% ça veut dire que le niveau de confiance est très bas.

edit j'avoue avoir eu la grosse flemme de lire l'article wiki sur la précision et comment la norme ISO 5725 la définissait... :D

Lien vers le commentaire

Créer un compte ou se connecter pour commenter

Vous devez être membre afin de pouvoir déposer un commentaire

Créer un compte

Créez un compte sur notre communauté. C’est facile !

Créer un nouveau compte

Se connecter

Vous avez déjà un compte ? Connectez-vous ici.

Connectez-vous maintenant
×
×
  • Créer...