mldlnl

The Most Efficient Way to Design AI

Project maintained by alihakimtaskiran Hosted on GitHub Pages — Theme by mattgraham

Documentation

1. What's the MLDLNL?

  MLDLNL is a tensorflow based high level API. It facilitates creating machine learning models(for now).

2.Installation

a.Requirements

-Python
     -Numpy
     -Tensorflow 1.x.x

b.Installation Steps

<p> You can run the module including into working directory. You can also install it into python3. After the installation, you can run the module without including into working directory.</p> Follow this steps for install the module(optional)

·Installation Steps for Linux

Open the terminal. After that
git clone https://github.com/alihakimtaskiran/mldlnl.git
cd "mldlnl"
sudo cp mldlnl.py /usr/lib/python[version]
If you haven't installed numpy, install the numpy with pip3 install numpy
If you haven't installed tensorflow 1.x, install the tensorflow 1.x with pip3 install tensorflow==1.15.0

Finally, you can use the module in python3 just one lines of code:import mldlnl

3.Tree of Module

mldlnl---|
         |
         |---LinReg()--|
         |             |--fit(x,y,lr=0.1,iter_no=80000,loss_fun="L2",lang="en")
         |             |--get_variables()
         |             |--calc(x)
         |             |--save(file_name)
         |             |--restore(file_name)
         |             |--restore_manually(weight,bias)
         |
         |
         |---MultiLinReg(n_of_params)--|
         |                             |--fit(x,y,lr=0.05,iter_no=70000,loss_fun="L2",lang="en")
         |                             |--get_variables()
         |                             |--calc(x)
         |                             |--save(file_name)
         |                             |--restore(file_name)                                   
         |
         |
         |---Perceptron(neurons=[1,1,1],activation_fun="tanh")--|
         |                                              |--fit(x,y,epochs=5,batch_size=200,lr=0.01,keep_prob=1.)
         |                                              |--calc(x,argmax=False)
         |                                              |--save(file)
         |                                              |--restore(file)
         |                                              |--test(x,y)
         |                                              |--properties()
         |
         |
         |---tools--|
         |          |--split_batch(x,batch_size)
         |
         |
         |---ExpReg()--|
         |             |--fit(x,y,lr=0.01,iter_no=50000)
         |             |--calc(x)
         |             |--get_variables()
         |             |--save(n_of_file)
         |             |--restore(n_of_file)
         |---CExpReg()--|
                        |--fit(x,y,lr=0.01,iter_no=50000)
                        |--calc(x)
                        |--get_variables()
                        |--save(n_of_file)
                        |--restore(n_of_file)

LinReg:
```
type:objectLinear Regression object. Use this object to create linear regression models.
```

MultiLinReg(n_of_params)
Multi Linear Regression object. Use this object to create multi linear regression models.</pre>

type:object

n_of_params is number of parameters.

fit(x,y,lr=0.1,iter_no=80000,loss_fun="L2",lang="en")

type:method fit() optimizes model with specific loss function. It uses tf.train.AdamOptimizer to find optimum weight and bias.x is input,y is output. lr is learning rate, it's default 0.1.iter_no is number of train step.loss_fun is a string represents loss function.It's default L2, you can also use L1 with "L1".

get_variables()

type:method The function exports variables and returns a tupleweight,bias)

calc(x)

type:method

Computes the output value with spesific input.

save(file_n)

type:method

Exports and saves parametrs into a file.file_n is a string takes name of file. Don't add directory into the string.

restore(file_n)

type:method

Imports and restores parameters from file.file_n is a string takes name of file. Don't add directory into the string.

Perceptron(neurons=[1,1,1],activation_fun="tanh")
Perceptron object utilizes creating multi layer perceptrons with specific activation function </pre>

type:object

neurons is a list(like [784,256,128,10]) to represent number of neuron per layer. The list should have at least 3 elements.activation_fun is specific activation function for perceptron.tanh,ReLU and sigmoid are supported activation functions. Activation function of last layer is softmax independed from specific activation function.

fit(x,y,epochs=5,batch_size=200,lr=0.01,keep_prob=1.)

type:method fit() utilizes train perceptron. It uses AdamOptimizer to optimize model.x is input data and y is output data to train perceptron. You don't need to split into batchs the data. Percepton object has internal batch splitting system.epochs is training epochs.batch_size is default set into 200.lr is default set into 0.01.keep_prob is probibilty of retained neurons after dropout

```
calc(x,argmax=False) 
```

type: method This function feed forwards an input value. You can compute the output of perceptron.x is input. argmax is a boolean. If it's True, the function returns index of maximum value of percpetron's output. If it's False, function returns output of perceptron.

```
save(file) 
```

type:method It exports parameters of perceptrons into a filefile is name of the file.

```
restore(file) 
```

type:method It imports parameters of perceptron from file. file is name of the file.

```
test(x,y) 
```

type:method Computes cross entropy loss and accuracy of spesific data. x is input and y is true output. Function returns loss,accuracy

```
properties() 
```

type:method Shows number of neurons in each layer and activation function. Function returns a tuple (neurons, activation function)

tools()
This class contains usefull tools for data science.</pre>

type:class

```
split_batch(x,batch_size) 
```

type:method This function facilates spliting dataset into batchs.x is input data, batch_size is size of batch, it's an integer.

ExpReg()

type:class
fit(x,y,lr=0.01,iter_no=50000)
```
type:method
```
It optimizes the model with dataset. x is dataset's x values and y is y values of dataset.lr is learning rate set as 0.01. iter_no is training steps for optimizer.
```
calc(x) 
```
```
type:method
```
Computes the output value with spesific input.
```
get_variables() 
```

type:method The function exports variables and returns a tupleweight,bias)

```
save(n_of_file) 
```

type:method

Exports and saves parametrs into a file.n_of_file is a string takes name of file. Don't add directory into the string.

```
restore(n_of_file) 
```

type:method

Imports and restores parameters from file.n_of_file is a string takes name of file. Don't add directory into the string.

CExpReg()

type:class
fit(x,y,lr=0.01,iter_no=50000)
```
type:method
```
It optimizes the model with dataset. x is dataset's x values and y is y values of dataset.lr is learning rate set as 0.01. iter_no is training steps for optimizer.
```
calc(x) 
```
```
type:method
```
Computes the output value with spesific input.
```
get_variables() 
```

type:method The function exports variables and returns a tupleweight,bias)

```
save(n_of_file) 
```

type:method

Exports and saves parametrs into a file.n_of_file is a string takes name of file. Don't add directory into the string.

```
restore(n_of_file) 
```

type:method

Imports and restores parameters from file.n_of_file is a string takes name of file. Don't add directory into the string.

mldlnl

Documentation

1. What's the MLDLNL?

2.Installation

a.Requirements

b.Installation Steps

·Installation Steps for Linux

3.Tree of Module

LinReg:

MultiLinReg(n_of_params)

Perceptron(neurons=[1,1,1],activation_fun="tanh")

tools()

ExpReg()

CExpReg()