Skip to content

Commit

Permalink
More changes to wording/formatting, fixing small errors
Browse files Browse the repository at this point in the history
  • Loading branch information
@Thelukepet
Thelukepet committed Aug 21, 2023
1 parent 7b42a17 commit 10b7b80
Showing 1 changed file with 38 additions and 36 deletions.
74 changes: 38 additions & 36 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -73,22 +73,22 @@ The user can decide the type of non-linearity function for certain specified lay
The current VGSL-Specs implementation supports the following layers:
## Overview Sheet

| **Layer** | **Spec** | **Example** | **Description** |
|--------------------|------------------------------------------------|------------------|----------------------------------------------------------------------|
| Input | `[batch, height, width, depth]` | `None,64,None,1` | Input layer with variable batch_size & width, depth of 1 channel |
| Output | `O(2\|1\|0)(l\|s)` | `O1s10` | Dense layer with a 1D sequence as with 10 output classes and softmax |
| Conv2D | `C(s\|t\|r\|e\|l\|m),<x>,<y>[<s_x>,<s_y>],<d>` | `Cr,3,3,1,1` | Conv2D layer with Relu, a 3x3 filter and 1,1 stride |
| Dense (FC) | `F(s\|t\|r\|l\|m)<d>` | `Fs64` | Dense layer with softmax and 64 units |
| LSTM | `L(f\|r)[s]<n>` | `Lf64` | Forward-only LSTM cell with 64 units |
| GRU | `G(f\|r)[s]<n>` | `Gr64` | Reverse-only GRU cell with 64 units |
| Bidirectional | `B(g\|l)<n>` | `Bl256` | Bidirectional layer wrapping a LSTM RNN with 256 units |
| BatchNormalization | `Bn` | `Bn` | BatchNormalization layer |
| MaxPooling2D | `Mp<x>,<y>,<s_x>,<s_y>` | `Mp2,2,1,1` | MaxPooling2D layer with 2,2 pool size and 1,1 strides |
| AvgPooling2D | `Ap<x>,<y>,<s_x>,<s_y>` | `Ap2,2,2,2` | AveragePooling2D layer with 2,2 pool size and 1,1 strides |
| Dropout | `D<rate>` | `Do25` | Dropout layer with `dropout` = 0.25 |
| ------------------ | ------------------------------------ |------------------|----------------------------------------------------------------------|
| ResidualBlock | `TODO` | | |
| CTCLayer | `TODO` | | |
| **Layer** | **Spec** | **Example** | **Description** |
|--------------------|------------------------------------------------|--------------------|------------------------------------------------------------------------|
| Input | `[batch, height, width, depth]` | `None,64,None,1` | Input layer with variable batch_size & width, depth of 1 channel |
| Output | `O(2\|1\|0)(l\|s)` | `O1s10` | Dense layer with a 1D sequence as with 10 output classes and softmax |
| Conv2D | `C(s\|t\|r\|e\|l\|m),<x>,<y>[<s_x>,<s_y>],<d>` | `Cr,3,3,64` | Conv2D layer with Relu, a 3x3 filter, 1x1 stride and 64 filters |
| Dense (FC) | `F(s\|t\|r\|l\|m)<d>` | `Fs64` | Dense layer with softmax and 64 units |
| LSTM | `L(f\|r)[s]<n>` | `Lf64` | Forward-only LSTM cell with 64 units |
| GRU | `G(f\|r)[s]<n>` | `Gr64` | Reverse-only GRU cell with 64 units |
| Bidirectional | `B(g\|l)<n>` | `Bl256` | Bidirectional layer wrapping a LSTM RNN with 256 units |
| BatchNormalization | `Bn` | `Bn` | BatchNormalization layer |
| MaxPooling2D | `Mp<x>,<y>,<s_x>,<s_y>` | `Mp2,2,1,1` | MaxPooling2D layer with 2x2 pool size and 1x1 strides |
| AvgPooling2D | `Ap<x>,<y>,<s_x>,<s_y>` | `Ap2,2,2,2` | AveragePooling2D layer with 2x2 pool size and 1x1 strides |
| Dropout | `D<rate>` | `Do25` | Dropout layer with `dropout` = 0.25 |
| ------------------ | ------------------------------------ | ------------------ | ---------------------------------------------------------------------- |
| ResidualBlock | `TODO` | | |
| CTCLayer | `TODO` | | |

## Layer Details
### Input:
Expand All @@ -107,37 +107,39 @@ _Creates a Dense layer with a 1D sequence as output with 10 classes and softmax_

### Conv2D:
<u>Spec</u>: **`C(s|t|r|e|l|m)<x>,<y>,[<s_x>,<s_y>,]<d>`** <br>
Convolves using a `x`,`y` window and `d` units. Optionally, the stride window can be set with (`s_x`, `s_y`) <br>
Example: `Cr3,3,1,1` <br>
_Creates a Conv2D layer with a Relu activation function a 3x3 filter and 1,1 stride (if s_x and s_y are not provided, set to (1,1) default)_
Convolves using a `x`,`y` window and `d` filters. Optionally, the stride window can be set with (`s_x`, `s_y`) <br>
Example_1: `Cr3,3,64` <br>
_Creates a Conv2D layer with a Relu activation function a 3x3 filter, 1x1 stride (if s_x and s_y are not provided, set to (1,1) default) and 64 filters_
Example_2: `Cr3,3,1,3,128` <br>
_Creates a Conv2D layer with a Relu activation function a 3x3 filter, 1x3 strides and 128 filters_
<br><br>

### **Dense (Fully-connected layer)**:
<u>Spec</u>: **`F(s|t|r|e|l|m)<d>`** <br>
Fully-connected(FC)) with `s|t|r|e|l|m` non-linearity and `d` outputs <br>
Fully-connected(FC)) with `s|t|r|e|l|m` non-linearity and `d` units <br>
Example: `Fs64` <br>
_Creates a FC layer with softmax non-linearity and 64 outputs_
_Creates a FC layer with softmax non-linearity and 64 units_
<br><br>

### **LSTM**:
<u>Spec</u>: **`L(f|r)[s]<n>`** <br>
LSTM cell running either forward-only (`f`) or reversed-only (`r`), with `n` outputs. Optionally, summarization of the output at the final step can be added by providing a Boolean `d` (corresponds with the return_sequences)<br>
LSTM cell running either forward-only (`f`) or reversed-only (`r`), with `n` units. Optionally, summarization of the output at the final step can be added by providing a Boolean `d` (corresponds with the return_sequences)<br>
Example: `Lf64` <br>
_Creates a forward-only LSTM cell with 64 outputs_
_Creates a forward-only LSTM cell with 64 units_
<br><br>

### **GRU**:
<u>Spec</u>: **`G(f|r)[s]<n>`** <br>
GRU cell running either forward-only (`f`) or reversed-only (`r`), with `n` outputs. Optionally, summarization of the output at the final step can be added by providing a Boolean `d` (corresponds with the return_sequences)<br>
GRU cell running either forward-only (`f`) or reversed-only (`r`), with `n` units. Optionally, summarization of the output at the final step can be added by providing a Boolean `d` (corresponds with the return_sequences)<br>
Example: `Gf64` <br>
_Creates a forward-only GRU cell with 64 outputs_
_Creates a forward-only GRU cell with 64 units_
<br><br>

### **Bidirectional**:
<u>Spec</u>: **`B(g|l)<n>`** <br>
Bidirectional layer which wraps either a LSTM (`l`) or GRU (`g`) RNN layer and runs it in both forward and backward directions. <br>
Bidirectional layer which wraps either a LSTM (`l`) or GRU (`g`) RNN layer and runs it in both forward and backward directions with `n` units. <br>
Example: `Bl256` <br>
_Creates a Bidirectional RNN layer using a LSTM Cell with 256 outputs_
_Creates a Bidirectional RNN layer using a LSTM Cell with 256 units_
<br><br>

### **BatchNormalization**:
Expand All @@ -149,24 +151,24 @@ Applies a transformation that maintains the mean output close to 0 and the outpu
<br><br>

### **MaxPooling2D**:
<u>Spec</u>: **`Mp<x>,<y>,<s_x>,<s_y>`** <br>
Downsampling technique along the height and width of a 2D image by taking the maximum value over the input window `x`,`y` and is shifted by strides along each dimension (`s_x`, `s_y`). `padding` is always set to "same".<br>
<u>Spec</u>: **`Mp<x>,<y>,<s_x>,<s_y>`** <br>
Example: `Mp2,2,2,2` <br>
_Creates a MaxPooling2D layer with pool size (2,2) and strides of (2,2)_
Example: `Mp2,2,2,2` <br>
_Creates a MaxPooling2D layer with pool size (2,2) and strides of (2,2)_
<br><br>

### **AvgPooling2D**:
<u>Spec</u>: **`Ap<x>,<y>,<s_x>,<s_y>`** <br>
Downsampling technique along the height and width of a 2D image by taking the average value over the input window x,y and is shifted by strides along each dimension (s_x, s_y). `padding` is always set to "same".<br>
<u>Spec</u>: **`Ap<x>,<y>,<s_x>,<s_y>`** <br>
Example: `Ap2,2,2,2` <br>
_Creates a AveragePooling2D layer with pool size (2,2) and strides of (2,2)_
Example: `Ap2,2,2,2` <br>
_Creates a AveragePooling2D layer with pool size (2,2) and strides of (2,2)_
<br><br>

### **Dropout**:
Regularization layer that randomly sets input units to 0 with a frequency of `rate` at each step during training time. Used to prevent overfitting.
<u>Spec</u>: **`D<rate>`** <br>
Example: `Do50` <br>
_Creates a Dropout layer with a dropout rate of 0.5 (`D`/100)_
Regularization layer that randomly sets input units to 0 with a frequency of `rate` at each step during training time. Used to prevent overfitting.
Example: `Do50` <br>
_Creates a Dropout layer with a dropout rate of 0.5 (`D`/100)_
<br><br>

---
Expand Down

0 comments on commit 10b7b80

Please sign in to comment.