Deep Learning || Neural Network Part 1

 Post installing Tensorflow, you are enabled with power of neural networks to perform deep learning.

TensorFlow is an open-source end-to-end platform for Machine Learning

1. First thing first:

import tensorflow as tf

import numpy as np

from tensorflow import keras

2. Load model configurations

model = tf.keras.Sequential([keras.layers.Dense(units=1, input_shape=[1])])

It has 1 layer, and that layer has 1 neuron, and the input shape to it is just 1 value.

3. Compiling neural network

The model uses the optimizer function to make another guess. Based on the loss function's result, it will try to minimize the loss. The model will repeat this for the number of epochs which you will see shortly.

But first, here's how we tell it to use `mean squared error` for the loss and `stochastic gradient descent` (sgd) for the optimizer. You don't need to understand the math for these yet, but you can see that they work!

model.compile(optimizer='sgd', loss='mean_squared_error')

4. Defining or creating data of 2 random arrays 

xs = np.array([-1.0, 0.0, 1.0, 2.0, 3.0, 4.0], dtype=float)

ys = np.array([-2.0, 1.0, 4.0, 7.0, 10.0, 13.0], dtype=float)

The algorithm will try and see the relationship between the 2 arrays and allow the prediction for any value which we input.

5. Fitting the model with 500 epochs

model.fit(xs, ys, epochs=500)

An epoch is a term used in machine learning and indicates the number of passes of the entire training dataset the machine learning algorithm has completed. Datasets are usually grouped into batches (especially when the amount of data is very large)

6. Predicting 

print(model.predict([10.0]))

You might have thought 31, right? But it ended up being a little over. 

Neural networks deal with probabilities, so given the data that we fed the NN with, it calculated that there is a very high probability that the relationship between X and Y is Y=3X+1, but with only 6 data points we can't know for sure. As a result, the result for 10 is very close to 31, but not necessarily 31.

Thanks,

Happy coding

Installing fbprophet

fbprophet is one of my favourite tools for forecasting. Now the problem is in the first step to install fbprophet. Sometimes it can be done via just "pip install fbprophet" in Anaconda prompt. It gives you problem as it might show many errors for dependent packages not being installed. In your Jupyter notebook it will still show "Module not found" etc.

I looked into the issue for long. I found multiple solutions to the problem on various portals but the one which will definitely gives you results is the below one:

First you can try with the below command on Anaconda Prompt. 

conda install -c conda-forge fbprophet

This might not work always but the below command will surely give you the results:

On Anaconda Prompt give the below commands

1. On Prompt install Ephem:   conda install -c anaconda ephem

2. Install Pystan:  conda install -c conda-forge pystan

3. Finally install Fbprophet:  conda install -c conda-forge fbprophet

This might take sometime for conda to download and install different packages but it will surely give you results.

Regards,

Power Bi or Tableau || Visualization

Visualization is nowadays one of the most important fields within analytics domain. To get handy insights and develop an understanding i.e. exploratory analysis it is best to do some quick visualization to see the big picture. And once we are done with data preprocessing, some cool visualizations help us in developing better understanding of our products, customer, geographies etc. Also now we have strong tools from Tableau to Power Bi which give us more than just visualization but also analytics within the tools. 

Lets start: You first need to download the free version for Power BI Desktop and start practicing. I will post some tutorials over next couple of weeks to bring you to speed with latest and robust features in Power Bi. You can also download Tableau Public for free. Although these free version do not have full features but they can give you a very good platform for training purposes.

Tableau Visuals

Power Bi Visual

While Tableau may look easy to use to some users, Power bi has strong capabilities in terms of modeling your data. Structuring your data maybe easier in Power Bi since it uses Excel like features. You may also use Alteryx for preprocessing and then send data to SQL server to load into either power bi or tableau or you can even download as excel or csv post Alteryx and then upload to Tableau or Power Bi. 

For calculations, you need to create calculated fields in Tableau, and measures in Power Bi. Measures might need some understanding of the M language. While both tools give shortcuts to create measures and calculated fields. But more than 90% of the time you go about creating custom measures and calculated fields. There are many custom visuals in Power Bi which give amazing visualization features such as Gantt charts, candle sticks etc.

Creating Measure in power bi

newSales = SUMX(Salestable,Salestable[Amount]*[newprice])

Calculated fields in Tableau

newSales = [amount]*[newprice]

While you need to give context filters within Dax measures in power bi. You can simply enable

add to context feature in filter in Tableau.

Both tools are amazing in their own way. You can chose one and get started with learning and once you have hands on experience,

you can easily get to another tool.

Thanks

Time Series Forecasting using fbprophet in Python

Hello friends. Today I am going to talk about time series forecasting using one of the best tools that we have in open source projects i.e. fbprophet. You can read more about it with documentation at the hyperlink. 'fbprophet' implements a procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality and on top of that gives flexibility for holiday effects.

So lets start, I have chosen a kaggle dataset to predict the sales of stores. You can probably download the dataset and practice along for a better understanding. 

Importing the necessary libraries: As always

But most importantly, it will ask you to install a few packages as 99% chances are, you wont have these packages installed. So you need to install in Ananconda prompt using pip install. For e.g.

Step 2: Plotting the data

plt.subplots(1, figsize = (16, 5))

plt.grid()

plt.title('Store 1 - All Sales')

for item in set(df_raw[df_raw['store'] == 1]['item']):

    plt.plot(df_raw[(df_raw['store'] == 1) & 

                    (df_raw['item'] == item)]['sales'].rolling(window=50).mean())

Step 3: Preparing the data into Fbprophet format

train_dataset = df_raw[(df_raw['item'] == 1) & df_raw['store'] == 1] 

train_dataset.reset_index(level=0, inplace=True)

train_dataset = train_dataset[['date', 'sales']]

train_dataset.columns = ["ds", "y"]

train_dataset = train_dataset.sample(5000)

prophet_basic = Prophet()

prophet_basic.fit(train_dataset)

Step 4: Forecasting 1 year ahead

future= prophet_basic.make_future_dataframe(periods=365)

future

Step 5: And here you go with the forecast

You can visualize the various components i.e. yearly and weekly trends

fig1 = prophet_basic.plot_components(forecast)

Visualize with even further additions to the plot.

So time series analysis is now simple and robust with fbprophet. You just need to have the right set of data.

Since we are talking a lot about visualization within data analytics, so maybe I will come up with a post on amazing Tableau visualizations next.

Thanks,