84 Matching Annotations
  1. Jan 2023
    1. There are no existing augmentation techniques that can correct adataset that has very poor diversity with respect to the testing data. All these augmenta-tion algorithms perform best under the assumption that the training data and testingdata are both drawn from the same distribution.

      Important border to where these can be applied!

    2. By taking a test image and augmenting it in thesame way as the training images, a more robust prediction can be derived.

      Interesting

  2. Jan 2022
    1. Conclusion

      It might be me, but I was missing a section on possible future directions and the drawbacks of using this model. Noticed that there was a lot of praise, but couldn't find a lot of critiques. (Or I might overseen it as I'm writing this at half 4 in the morning.)

      Normally, you could find the published reviews on OpenReview with the discussed critique (see: https://papers.nips.cc/paper/2016/file/fb87582825f9d28a8d42c5e5e5e8b23d-Reviews.html which I used for another repro course), but when googling on this paper, I couldn't find it :(

      Either way, here's the OR link to the current paperhttps://openreview.net/forum?id=FOL6UA7JYtk)

  3. Nov 2021
    1. gure 26.8

      The trader adopting this strategy believes in some sort of range, since once the stop price of the buy stop is reached, I believe that it triggers a limit buy order below the market or at least at the market

    2. At extremely high levels of leverage, the concept of aggressiveness and conservativeness with respect to the price on entry or exit di-minishes rapidly

      Because with a small capital, one can create a large margin position => sheer capital invested is no longer significant due to leverage

    3. gUre 25.4

      Si ca dépasse le niveau du buystop, alors on déclenche un ordre d'achat proche du support

    4. It should also be noted that the longer it takes for a consolidation to unfold, the larger will be the number of potential buy stop orders that lie above the resistance zone and sell stop orders that populate just below the support zone. This explains why larger consolidations tend to produce stronger and more persistent breakouts

      buy stop orders place limit orders at the support, soif a lot of stop buy orders are triggered ,this places manylimit orders at support, and allows price to rebound strongly enough to pierce through the resistance

    5. total short sales

      Meaning : by the public (under-informed) and by the well-informed

    6. Cash/asset ratio

      En gros :

      • Cash/asset ratio grand => marché descendant
      • Cash/asset ratio petit => marché ascendant
    7. we could not determine which stock was more volatile since the maximum amount of price change over equal durations were the same for both
      • What about comparing over smaller time periods, if we do this, we'll clearly see that there are differences ?
      • How do we pick the timeframe ?
    8. exponential or parabolic

      Isn't it parabolic in t his example, since y'' = cte ?

    9. figure

      Mom >0 => fixed momentum

    10. Third‐Order Change

      ROC of ROC of PC

    11. Second‐Order Change

      ROC of PC

    12. First‐Order Change

      PC

    13. gure 21.3
      • Arithmetic change of price for stock A

      vs.

      • Geometric progression for B
    14. price change

      Analogy with mechanics :

      • PC == speed
      • ROC of PC == acceleration
    15. Centered moving average

      SEE page 435

    16. This is because the bands are adjusted with respect to actual price action over a certain period of observation

      Period of observation which is the band's periodicity ?

    17. confirmation in the near or front month futures contract

      Crypto -> expiration des contrats futurs ?

    18. igure 18.7

      Dès qu'il y a 5 waves, elles sont dans la direction de la wave une échelle au-dessus, ce qui dans notre cas est descendant, d'où la (1),(2),(3),(4),(5) descendant ici !

    19. gure 18.5

      SEE paper version

    20. MACD

      See page 258

    21. ounding tops and bottoms

      U shape

    22. gets may also be obtained based on the parallelogram itself

      cf. Fig. 13.37, where we project the objective price targets based on the parallelogram and not on the pole !

    23. subside

      diminuer/se calmer

    24. igure 11.22

      MACD represented as a histogram instead of a curve/line as of usually. But we see that each time the MACD signal line crosses the MACD (represented as a histogram), the MACD histogram is equal to zero !!

    25. gure 8.19

      The histogram in the MACD window (window above the MACD histogram window) and the histogram in the MACD histogram window are the same histogram, just with a different scaling !

    26. no lag with respect to price

      But we need to know future information since it is located in the middle of N bars => we need to know N/2 bars in the future. That's why there's no lag !!

    Tags

    Annotators

  4. Oct 2021
    1. Figure 10.69

      SEE paper version

    2. logarithmic‐ or arithmetic‐scaled chart is employed

      For static, both work since it only depends on the price, so we can determine it numerically. However, since fan lines are dynamic, they depend on price AND time, so we can't just determine the levels numerically.

    3. the low of the day

      Coquille dans le graphique ??

    4. Figure 10.36
      • Avec l'échelle linéaire, deux graduations dont la différence vaut 10 sont à distance constante.
      • Avec l'échelle logarithmique, deux graduations dont le rapport vaut 10 sont à distance constante
    5. Figure 10.35

      Même remarque que pour la figure précédente !

    6. Figure 10.34

      Mêmes calculs => mêmes niveaux (38,2, 50,0, 61,8, ...) mais l'échelle (ie. l'axe des y) est log au lieu d'être linéaire.

    7. Figure 10.27

      SEE âge 380 of the paper version

    8. Fibonacci downside extension level
      • Use peak as base when downside Fibo
      • Use trough as base when upside Fibo
    9. Figure 10.26

      SEE page 377 of paper version for detailed graphs

      We put 100% between A and B because we look at the extension level beyond B so x% with x>=100%

      Also, we start hte Fibonacci extension at point A instead of B even though B is located earlier in time than A since A is a trough and B is a peak and it would be highly unconventional to pull Fibonacci levels from B to A

    10. 423.6

      (Phi)^3

      Or divide a number by 3 to the left

    11. 261.8

      (Phi)^2

    12. 161.8

      Phi

    13. 127.2

      sqrt(Phi)

    14. 38.2

      1/Phi FAUX car Phi = 1 + (1/Phi)

      Obtained by dividing a sequence number by the sequence number 2 spots to the right : (Fn) / (F{n-2})

    15. 23.6

      (2/Phi)-1

      Also (Fn) / (F{n-3})

    16. 61.8

      Phi - 1

    17. 78.6

      sqrt(1/Phi)

    18. multiply

    19. Figure 9.101

      Relative since we compare ratios (the y axis is a % axis !)

    20. Figure 9.93

      Peaks and troughs are not trending in the same direction in expanding or contracting formations !

    21. This is because volume oscillators are constructed to take into account price

      Explanation of why volume indicators are analyzed using standard and reverse divergences instead of volume, open interest and ATR.

    22. The rules for interpreting divergence are markedly different when it comes to vol-ume bar action, open interest, and average true range (ATR)

      Remember this when reading the next sections on divergence !! Not the same interpretation at all as for the other supporting data series !!

    23. nter Wave Cycle/Degree Divergence

      Une divergence "contenue" dans une autre à une larger wave degree

    24. nter peak or trough Double Divergence

      Deux divergences qui se suivent

    25. thE hAnDbook of tEchnicAL AnALySiS292In general terms, basic reverse divergence occurs when: ■ Only price, and not the supporting data series, is making equal or lower peaks (based on adjacent peak to peak analysis) ■ Only price, and not the supporting data series, is making equal or higher troughs (based on adjacent trough to trough analysis).

      cf. last note

    26. 1. Price is making higher peaks while the oscillator is making lower peaks 2. Price is making higher peaks while the oscillator is making equal peaks 3. Price is making equal peaks while the oscillator is making lower peaks

      Here, the price only makes higher or equal peaks -> lower peaks is for reverse divergence -> called "reverse divergence" since we start by giving the direction of the oscillator (only up or equal) and subsequently we give the direction of the price.

    27. peak and trough analysis

      cf. p. 178 sur bearish divergence et bearish confirmation

    28. Defining Direction of the Current Larger Trend

      On double de 9 à 18 car on distingue le peak-to-peak du trough-to-trough -> 2 fois plus

    29. collapse back to six

      Car on ne distingue plus le peak-to-peak du trough-to-trough -> 2 fois moins

    30. moving in opposing directions

      cf. version papier

    31. adjacent peaks or troughs

      In this case, the drawing on the paper version is an upper trend. Cf. Fig. 9.1

    32. George Lane

      Stochastic oscillator (cf. p.259 of chap. 8)

    33. True Range Action of the Daily Chart of Gold. Source: MetaTrader 4

      L'ATR est très "non smooth" car on n'utilise qu'une lookback period de 1. Si on faisait la moyenne mobile sur plus d'une période, il serait plus smooth.

    34. Intrinsic

      Intrinsic in the sense of chap. 4

      E.g. a head and shoulder pattern is bearish intrinsically, but it can happen in an uptrend and in this case, intrinsic sentiment and the trend disagree !

    35. Rising and declin-ing volume provide useful information about the level of interest or participation at various prices and indicate whether an ongoing trend is bearish or bullish

      Cf. ce qui a été dit au chap 4 sur l'intérêt des participants du marché et l'impact sur le volume.

    36. Volume Blow‐Offs

      Quand ca commence à omnter, il faut que le volume monte aussi, sinon on considère que ce n'est pas un uptrend, mais juste une correction. Pareil opur quand ça descend, il faut du volume pour accompagner la baisse :

    37. volume declining during a consolidation

      This shows that declining volume during consolidation is also a thing to take into account

    38. reversal of the existing trend

      Hence if the trend is a retracment, we will see the bullish trend resume shortly

    39. Volume

      See page 113 for the relationship between volume and price action (volume declines during consolidation phases).

    40. Note that for bullish divergence in volume, falling prices must be accompanied by decreasing volume.

      A l'inverse d'autres indicateurs comme le MACD ou le RSI

    41. In markets where the trading is continuous

      cf. crypto

    42. The durations between active trading sessions are longer

      Does not concern crypto since the market is active 24/7

    43. industrial or transportation

      Industrial and transportation averages !

    44. give earlier trend change signals

      Uptrend lines are violated sooner since logarithm tends to rapprocher les points éloignés arithmétiquement (de manière non linéaire, d'où la cassure de la ligne de tendance plus tôt)

    45. The pricing of all known information need not be instantaneous or be driven by rational participants. There is also no requirement that all participants always act on all information all of the time, or that they react in the same manner.

      Difference with EMH

    46. he mechanism by which information is known to the market is that of actual participation via capital injection.

      This means that any information is accounted for by capital injection

    47. Markovian condition

      In the sense that the possibilities for future price are given by current price, but there is then no way to predict which possibilty will be chosen. E.g. random walk in Z^2, if we're wurrently at a certian point, we have 4 possibilities for next move which are equiprobable

    48. Therefore, they cannot impact future prices

      Past price cannot influence future price, hence TA is useless as it uses past info to predict future price mvts.

    49. nstantaneously

      cf. remark on "Gradually" in Fig.1.28

    50. Gradually

      Gradually since on the graph we don't have a vertical spike like in the EMH case. This means that participants do not all react at the same time.

    51. Efficient Market Adjusting to New Information

      No insider information that would be discounted by the price, nor irrational behaviours like selling on a bullish news

    Tags

    Annotators

  5. Aug 2021
    1. Now some people may just say “Pffft… just get up earlier and do it”. Clearly those people don’t understand how much I love to get up as late as I can afford in the mornings, those moments between asleep and ‘Fuck, I have to go to work’ are few, and precious my friends.

      I resemble this remark...

  6. Oct 2020
  7. Dec 2017
    1. Table 1

      This paper https://arxiv.org/abs/1005.2012 of 2010 proves $\tilde{O}(\frac{Mm}{\epsilon^2})$ in the primal case. It should be in the related work