Taking classes at MPS Yokohama
I am struggling because I am a beginner in both programming and mathematics. (I will check Markdown notation and TeX notation and rewrite it later)
Ultimately, I would like to customize the appropriate Deep Learning library (tensorflow) and utilize it for work (advertising operation tool) and play (home appliance operation from sensing ??).
MPS Lecture Slides Lecture video
@ hi_saito's Great Summary Note (I'm sorry if there is a typographical error (?) Or my lack of understanding)
2016/5/25 "Unknown list"
1.1.1 2nd Slide p36 Great note p14
The weight W is written in two ways, the matrix W and the vector W (bold letters) translocation (?) T, and I think that both are the same as the resulting numbers, but I wonder if there is something particular about it. ?? → 2nd Lecture video 1h23min ~ 1h42min is the explanation The weights v00 and v01 required to derive z0 are two y0 and y1, so they are vectors, aren't they? Is suspicious. If it's a 2-by-1 matrix, it's a special vector, so I understand it now. I wonder if it is a device to make it a form that can be processed by numpy.dot (w, x).
1.1.1 Axes3D in the module mpl_toolkits.mplot3d How to draw here
W,B = np.meshgrid(w,b)
Z = se(0, sigmoid(W,1,B))
fig = plt.figure()
ax = Axes3D(fig)
ax.plot_wireframe(W,B,Z)```
2. What are you trying to achieve with numpy mesh grid?
3.```if __name__ == '__main__':```
I'm embarrassed not to understand. .. .. The function you called? class? If is main, do you run it? ?? What do you mean?
4.numpy.vectorize () Why use vectorize to create an array?
5.self.w [0] [0] What are the elements of this matrix w?
→ Maybe understanding.
#### **`w = [[3.0,],]`**
```0,],]
b = [[1.0,],]
alpha = 0.1
w=np.concatenate((w,b),axis=1)
w[0][0] += alpha```
Since the array [row] [column] of numpy, w [0] [0] is the 1st row and 1st column of the array w.
What kind of array w is is a 1-by-2 matrix called [3.0 1.0] because b is included in the column direction.
So, what is this 3.0? The bitflip problem is that the input is one-dimensional with 1 or 0 and the output is also one-dimensional with 1 or 0.
There is one weight w for the input. Its weight is 3.0
This time, I want to increase or decrease this w by alpha depending on the result, so I take out 3.0 in the 1st row and 1st column with w [0] [0].
I'm adding or subtracting alpha minutes.
7. About the sigmoid function
Since 1 is divided by 1 + "positive number (or 0)", I understand that the maximum is 1 and the minimum is a number as close to 0 as possible.
What kind of number is this "positive number (or 0)" of e to the -αu power (a> 0)? 1 / e to the αu power
I wonder what the number is ... (Since u is Wx + b, there is a possibility that it may be negative. Then, 1 / e to the au power becomes e to the au power, and the larger a is, the more rapidly the sigmoid Does the result of the function approach 0?)
8. How to use f (u) in python
9. How to calculate the derivative by hand
10.
Recommended Posts