<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Erica Gabriel on Medium]]></title>
        <description><![CDATA[Stories by Erica Gabriel on Medium]]></description>
        <link>https://medium.com/@gabriel.erica3?source=rss-6f1460ce25f9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*zLzz3Q4tiz_AmdVIvn7CGQ.jpeg</url>
            <title>Stories by Erica Gabriel on Medium</title>
            <link>https://medium.com/@gabriel.erica3?source=rss-6f1460ce25f9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 19:27:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gabriel.erica3/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How to Treat Overfitting in Convolutional Neural Networks]]></title>
            <link>https://medium.com/analytics-vidhya/how-to-treat-overfitting-in-convolutional-neural-networks-d6e74a0a479?source=rss-6f1460ce25f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/d6e74a0a479</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Erica Gabriel]]></dc:creator>
            <pubDate>Mon, 07 Sep 2020 13:53:36 GMT</pubDate>
            <atom:updated>2020-09-18T03:48:35.274Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Overfitting or high variance in machine learning models occurs when the accuracy of your training dataset, the dataset used to “teach” the model, is greater than your testing accuracy. In terms of ‘loss’, overfitting reveals itself when your model has a low error in the training set and a higher error in the testing set. You can identify this visually by plotting your loss and accuracy metrics and seeing where the performance metrics converge for both datasets.</p><p>Overfitting indicates that your model is too complex for the problem that it is solving, i.e. your model has too many features in the case of regression models and ensemble learning, filters in the case of Convolutional Neural Networks, and layers in the case of overall Deep Learning Models. This causes your model to know the example data well, but perform poorly against any new data.</p><p>This is annoying but can be resolved through tuning your hyperparameters, but first, let’s start by making sure our data is divided into well-proportioned sets.</p><h3>Splitting the Data</h3><p>For a deep learning model, I recommend having 3 datasets: training, validation, and testing. The validation set should be used to fine-tune your model until you’re satisfied with its performance, then switch to the testing data to train the best version of your model. First, we’ll import the necessary library:</p><pre>from sklearn.model_selection import train_test_split</pre><p>Now let’s talk proportions. My ideal ratio is 70/10/20, meaning the training set should be made up of ~70% of your data, then devote 10% to the validation set, and 20% to the test set, like so,</p><pre># Create the Validation Dataset <br>Xtrain, Xval, ytrain, yval = train_test_split(train_images, train_labels_final, train_size=0.9, test_size=0.1, random_state=42) </pre><pre># Create the Test and Final Training Datasets Xtrain, Xtest, ytrain, ytest = train_test_split(Xtrain, ytrain, train_size=0.78, random_state=42)</pre><p>You will need to perform <em>two train_test_split()</em> function calls. The first call is done on the initial training set of images and labels to form the validation set. We’ll call the parameters random_state to keep consistency in results when running the function, and test_size to note that we want the size of our validation set to be 10% of the training data, and train_size to set it equal to the remaining percentage of data to be 90%.</p><p>This can be omitted by default as python is smart enough to do the math. The variables Xval and yval refer to our validation images and labels. On the second call, we will generate our testing dataset from our newly formed training data Xtrain and ytrain. We will repeat the above, but this time we will set the newest training set to be 78% of the previous and assign the newest dataset to the same variable as the previous for consistency. Meanwhile, we will assign the testing data to Xtest for the test images and test for the label data.</p><p>Now we’re ready to begin modeling. Refer to my <a href="https://medium.com/analytics-vidhya/computer-vision-and-pneumonia-detection-part-1-technical-4e3592de208b?source=friends_link&amp;sk=665a349b71fc55dce132f6a28cfa5ef9">previous blog</a> to get a deep dive into the initial CNN setup. We will start on the second model assuming our first turned out like the image above. We will use the techniques below:</p><ol><li><strong>Regularization</strong></li><li><strong>Weight Initialization</strong></li><li><strong>Dropout Regularization</strong></li><li><strong>Weight Constraints</strong></li><li><strong>Other</strong></li></ol><h3>Regularization</h3><p><a href="https://www.analyticsvidhya.com/blog/2018/04/fundamentals-deep-learning-regularization-techniques/">Regularization</a> optimizes a model by penalizing complex models, therefore minimizing loss and complexity. Thus this forces our neural network to be simpler. Here we will use an<em> L2 regularizer</em>, as it is the most common and is more stable than an <em>L1 regularizer</em>. Here we’ll add a regularizer to the second and third layers of our network with a learning rate (lr) of 0.01.</p><pre># Hidden Layer 1 <br>model2.add(layers.Conv2D(64, (4, 4), activation=&#39;relu&#39;, kernel_regularizer=regularizers.l2(l=0.01))) <br>model2.add(layers.MaxPooling2D((2, 2))) </pre><pre># Hidden Layer 2 <br>model2.add(layers.Conv2D(128, (3, 3), activation=&#39;relu&#39;, kernel_regularizer=regularizers.l2(l=0.01))) model2.add(layers.MaxPooling2D((2,2)))</pre><h3>Weight Initialization</h3><p>Weight initialization sets up the weights vector for all neurons for the first time before the training process begins. Choosing the correct weights is crucial because we want to get as close as possible to the global minimum of our cost function in an adequate amount of time. In this iteration of our model we will use a <em>He initialization</em>:</p><pre># Input Layer of the 3rd Model <br>model3.add(layers.Conv2D(32, (3, 3), activation=&#39;relu&#39;, kernel_initializer=&#39;he_normal&#39;, input_shape=(96, 96, 3)))</pre><h3>Dropout Regularization</h3><p>Dropout regularization ignores a random subset of units in a layer while setting their weights to zero during that phase of training.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*MlYbJzwH43i0qLv4.png" /></figure><p>The ideal rate for the input and hidden layers is 0.4, and the ideal rate for the output layer is 0.2. See below:</p><pre>random.seed(123) # Establish Consistency in results <br>model4 = Sequential() # Instantiate the 4th Model model4.add(layers.Conv2D(32, (3, 3), activation=&#39;relu&#39;, input_shape=(96, 96, 3))) <br>model4.add(layers.MaxPooling2D((2, 2))) model4.add(Dropout(0.4)) model4.add(layers.Conv2D(64, (4, 4), activation=&#39;relu&#39;)) model4.add(layers.MaxPooling2D((2, 2))) model4.add(Dropout(0.4)) </pre><pre># Flattening- Convert 2D matrix to a 1D vector model4.add(layers.Flatten()) <br>model4.add(layers.Dense(512, activation = &#39;relu&#39;)) model4.add(Dropout(0.2)) <br>model4.add(layers.Dense(1, activation=&#39;sigmoid&#39;))</pre><h3>Weight Constraints</h3><p>A weight constraint checks the size of the network weights and rescales them if the size exceeds a predefined limit. The weight constraint works as required. Below we are using the constraint <em>unit_norm</em>, which forces the weights to have a magnitude of 1.0.</p><pre>model5.add(layers.Conv2D(32, (3, 3), activation=&#39;relu&#39;, kernel_constraint=unit_norm(), input_shape=(96, 96, 3)))</pre><h3>Other</h3><p>If all else fails, you can increase the size of your <strong>training set</strong> by generating more data. Here’s how to do so without altering the images:</p><pre>datagen = ImageDataGenerator(rotation_range = 0, width_shift_range = 0, height_shift_range = 0, rescale = None, shear_range = 0, zoom_range = 0, horizontal_flip = False, fill_mode = &#39;nearest&#39;)</pre><p>Another route is to increase the resolution of all of the photos by increasing the size. You can do this by calling new image data generators for the train, validation, and test datasets. See below how I increased the dimensions of the photos from (96 x 96) to (128 x 128):</p><pre># Import the Original Training Dataset <br>train_gen2 = ImageDataGenerator(rescale=1./255).flow_from_directory(train_dir, target_size=(128,128), batch_size=15200) </pre><pre># Import the Original Validation Dataset <br>val_gen2 = ImageDataGenerator(rescale=1./255).flow_from_directory(val_dir, target_size=(128,128), batch_size=16)</pre><pre># Import the Original Testing Dataset <br>test_gen2 = ImageDataGenerator(rescale=1./255).flow_from_directory(test_dir, target_size=(128,128), batch_size=624)</pre><p><strong>Erica Gabriel</strong></p><p>You can also read this article on our Mobile APP</p><p><em>Originally published at </em><a href="https://www.analyticsvidhya.com/blog/2020/09/overfitting-in-cnn-show-to-treat-overfitting-in-convolutional-neural-networks/"><em>https://www.analyticsvidhya.com</em></a><em> on September 7, 2020.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d6e74a0a479" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/how-to-treat-overfitting-in-convolutional-neural-networks-d6e74a0a479">How to Treat Overfitting in Convolutional Neural Networks</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[X-Ray Colorization and Neural Networks]]></title>
            <link>https://medium.com/@gabriel.erica3/x-ray-colorization-and-neural-networks-2cdc578a1d2f?source=rss-6f1460ce25f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/2cdc578a1d2f</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[medicine]]></category>
            <category><![CDATA[x-rays]]></category>
            <dc:creator><![CDATA[Erica Gabriel]]></dc:creator>
            <pubDate>Thu, 03 Sep 2020 06:10:58 GMT</pubDate>
            <atom:updated>2020-09-03T06:10:58.015Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/896/1*SPEADXmJWr0hGC4hVDEXAg.png" /><figcaption>Colorized x-ray image of a patient with pneumonia.</figcaption></figure><p>Why aren’t X-rays in color? The question that’s been on my mind for weeks while working to classify patients as having pneumonia or normal x-ray images from the <a href="https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia">Kaggle dataset</a>. To me it seems to be a no brainer. An upgrade comparable to when television and movies went from black and white to color. According to MedlinePlus,</p><blockquote>“The images show the parts of your body in different shades of black and white. This is because different tissues absorb different amounts of radiation. Calcium in bones absorbs x-rays the most, so bones look white. Fat and other soft tissues absorb less and look gray. Air absorbs the least, so lungs look black.”</blockquote><p>This is apparent in the images below. The current method could lead to certain features and abnormalities fading into the background.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qe2z6Fb8DyRSDF-Q3HWXiA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fENkfniD3DguT9z5UZ62IA.jpeg" /><figcaption>Two x-ray images taken from the training dataset.</figcaption></figure><p>Can you tell which patient has pneumonia? Can you accurately identify the abnormalities if you do not have a background as a radiologist? How can I augment these images so that even a 5 year old can point out the inflammation in the lungs? Colorize them. Through some soul searching, I stumbled upon the <a href="https://github.com/jantic/DeOldify/blob/master/README.md">DeOldify</a> repo, courtesy of <a href="https://twitter.com/citnaj"><strong>Jason Antic</strong></a><strong>. </strong>I cloned the repo and converted all 15,864 x-ray images I generated to color.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N1pc-LNhTQnOqgMvgfJWaQ.png" /><figcaption>A side by side view of an original x-ray scan vs. the colorized version of patient with bacterial pneumonia.</figcaption></figure><p>The grayscale dataset was more computationally expensive to build and train. The models took an average of 30 minutes to an hour to run, doing only 22 epochs. The grayscale models caused me to use nearly all of my local and google drive storage. The final and best model returned a validation accuracy of 86.3% and a validation loss of 0.000985.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/386/1*PhChUGn9n2zbou7pa0HQuw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/386/1*i2hDKNTSoBcJhrCLPqsZog.png" /><figcaption>Loss vs. Epoch and Accuracy vs. Epoch Plots for the grayscale dataset.</figcaption></figure><p>The colorized models ran for 100 epochs and took approximately 1 minute to run. The colorized dataset overall did not require much tuning or massaging compared to the grayscale dataset in order to achieve a low bias and low variance model. This set had a validation loss of 0.408 and a validation accuracy of 99.9%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/398/1*aaugfs3C4f0hrRi5W5NEDg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/392/1*zM7c29RdB6_2ChzzrrHoIQ.png" /><figcaption>Loss vs. Epoch and Accuracy vs. Epoch Plots for the colorized dataset.</figcaption></figure><p>The above is a side by side display of a patient with bacterial pneumonia. From the colorized x-ray, we are able to better decipher between the borders of the lung and the soft tissues inside of the body (the red areas shown in the augmented radiograph). This way we are able to determine where the mass or fluid accumulation occurs. We are also able to get an idea of the race of the patient, which is data that can be collected for clinical research for tracking trends and gathering insights that can lead to preventative care. If computer vision is going to be considered in detection and early diagnosis of pneumonia and similar illnesses, then research funding should be set aside to fund hardware and software developments that produce colorized and perhaps even 3D image scans.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2cdc578a1d2f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Computer Vision and Pneumonia Detection Part 2: Analysis]]></title>
            <link>https://medium.com/@gabriel.erica3/computer-vision-and-pneumonia-detection-part-2-analysis-6eec56d2cc69?source=rss-6f1460ce25f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/6eec56d2cc69</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[health]]></category>
            <category><![CDATA[covid19]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[computer-vision]]></category>
            <dc:creator><![CDATA[Erica Gabriel]]></dc:creator>
            <pubDate>Tue, 11 Aug 2020 00:04:59 GMT</pubDate>
            <atom:updated>2020-08-11T14:53:07.523Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/625/1*kyJ88BU3JAZP8U4pLrETGw.png" /><figcaption>Chest X-rays depicting the lungs of pediatric patients. The areas annotated in blue show evidence of damage due to pneumonia.</figcaption></figure><p>In my previous <a href="https://medium.com/@gabriel.erica3/computer-vision-and-pneumonia-detection-part-1-technical-4e3592de208b">post</a> I built a classifier to train a machine decipher between the chest x-rays of pediatric patients with “normal” lungs and lungs showing signs of pneumonia damage. In this part of the blog we will get into some exploratory data analysis. X-rays are the top and most common method of detecting pneumonia worldwide. Why, you ask?</p><ul><li>X-rays are a fast way to view internal organs &amp; and air-filled parts</li><li>They are inexpensive (For Medical professionals)</li><li>There’s a fast turnaround time on results</li></ul><p>However, the downsides of X-rays are that:</p><ul><li>Images can be difficult decipher</li><li>It’s not possible to detect the type of bacteria or the source from a scan</li><li>They can be costly for patients → <strong>with</strong> health insurance: <strong>$0 — $50</strong> &amp; <strong>without</strong> health insurance: <strong>$200 — $400</strong></li><li>Minuscule exposure to radiation</li></ul><p>It’s also important to note that the cost of x-rays varies per state. Why? <strong><em>What factors contribute to the fluctuating costs of x-ray scans nationwide?</em></strong> The Guangzhou Women and Children’s Medical patients were given chest x-rays as a part of their normal treatment. <strong>Being that pneumonia is recognized by the World Health Organization as one of the top 10 causes of death worldwide, why isn’t checking lung health a part of routine care worldwide?</strong> This led me to go down a rabbit hole to answer the question: <strong><em>What other (viral) diseases can be detected via chest x-ray?</em></strong> Low and behold, COVID-19, came up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/592/1*N6P5RcxGHP8vd_HuJmdXpw.png" /><figcaption>CDC table depicting the amount of COVID-19, Pneumonia, and Influenza in the United State since February 2020.</figcaption></figure><p>The CDC screenshot above shows the amount of COVID-19, Pneumonia, and Flu deaths since February 2020 in the United States. The chart reveals that 44% of people who died from COVID-19 had pneumonia. One of the long term effects of COVID is pneumonia induced lung injuries in both lungs. The current method of testing for COVID-19 is a nose swab, which takes about 2–3 days to get results because the samples often have to be sent to a lab. Another component to the current method of testing is that a negative result only indicates that the patient did not have COVID-19 <strong>AT THE TIME OF TESTING</strong>. Given the correlation between the two illnesses,<strong><em> why aren’t x-rays being considered as a means for COVID-19 detection?</em></strong> <strong><em>If adopted as a means of COVID-19 detection, would they be offered free of charge, similar to the viral test? Should testing for the top 10 deadliest diseases be free, or at an absolute minimum income based?</em></strong></p><p>After straining looking through literally thousands of x-rays, I wonder how much easier it would be to interpret the x-rays if they were in color. If you can tell from the leading image in the article that it can be pretty difficult to distinguish between a normal x-ray and an x-ray of an individual with viral pneumonia. <strong><em>Why aren’t x-ray scans taken in color?</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/444/1*hcazyP70ZcVMSvzBm9l37A.gif" /><figcaption>3D color scan of am ankle by MARS Bioimaging.</figcaption></figure><p>The above is a 3D color scan of an ankle by New Zealand company, <a href="https://www.marsbioimaging.com/">MARS Bioimaging </a>. The MARS scanner can distinguish between bone, muscle, and fat. The scan was taken in 2018. <strong><em>What is stopping the rest of the world from adapting 3D color scans? What impact would this method have on the accuracy and delivery of results? How soon could internal lung damage be detected if this method of scanning were to be adopted?</em></strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6eec56d2cc69" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Computer Vision and Pneumonia Detection Part 1: Technical]]></title>
            <link>https://medium.com/analytics-vidhya/computer-vision-and-pneumonia-detection-part-1-technical-4e3592de208b?source=rss-6f1460ce25f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e3592de208b</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[health]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[healthcare]]></category>
            <dc:creator><![CDATA[Erica Gabriel]]></dc:creator>
            <pubDate>Mon, 10 Aug 2020 19:40:06 GMT</pubDate>
            <atom:updated>2020-08-19T14:49:05.464Z</atom:updated>
            <content:encoded><![CDATA[<p>Recently, I tasked myself with analyzing chest x-ray images to determine whether or not a patient has pneumonia. The x-rays were provided by the Guangzhou Women and Children’s Medical Center in Guangzhou, China, on patients ranging from ages 1–5 years old. The image set contains images of “Normal”, “Viral” &amp; “Bacterial” Pneumonia, classified into 2 categories: “Normal”and “Pneumonia”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/625/1*kyJ88BU3JAZP8U4pLrETGw.png" /><figcaption>Chest X-rays depicting the lungs of pediatric patients. The areas annotated in blue show evidence of damage due to pneumonia.</figcaption></figure><p>The dataset was provided in ideal conditions, free of poor quality scans. Before building the model, I checked the training data for class imbalance. There were <strong>1341 Normal</strong> scans and <strong>3875 Pneumonia</strong> scans. Killing 2 birds with one stone, I added 2534 additional images, and augmented them to reflect real world scenarios- blurry, distorted images, shifted/rotated images, and images that are zoomed in and out of focus.</p><pre># Set criteria for image augmentation<br>datagen = ImageDataGenerator(<br>            rotation_range = 40,<br>            width_shift_range = 0.2,<br>            height_shift_range = 0.2,<br>            rescale = 1./255,<br>            shear_range = 0.2,<br>            zoom_range = 0.2,<br>            horizontal_flip = True,<br>            fill_mode = &#39;nearest&#39;)</pre><pre># Select an image from the dataset<br>img = load_img(&#39;chest_xray/train/NORMAL/IM-0757-0001.jpeg&#39;)</pre><pre># Convert image to an array<br>img_array = img_to_array(img)</pre><pre># Reshape the array to a (1 X n X m X 3) array<br>img_array = img_array.reshape((1,) + img_array.shape)</pre><pre># Set the path where the updated images should be saved<br>norm_dir = &#39;chest_xray/train/NORMAL/&#39;</pre><pre># Add 2534 photos to the directory in batches of 150<br>count = 0<br>for batch in datagen.flow(img_array, batch_size=150, save_to_dir=norm_dir, save_prefix=&#39;IM&#39;, save_format=&#39;jpeg&#39;):<br>    count +=1<br>    if count == 2534:<br>        break<br>        <br>print(&#39;2534 images have been generated at&#39;, norm_dir)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/320/1*KkgzbB1wm-lFVB-h8YhVTw.png" /><figcaption>Sample augmented image.</figcaption></figure><p>Once the images were added, I reduced the size of the images to 64 x 64, so that the sizes would be uniform and less computationally expensive to input into the model, and save the updated images to the original file directory. Using the next() function, I iterated through the image files and returned the images and labels that have been stored in the variables “__images” and “__labels”, then I normalized the images by dividing the pixels by 255.</p><pre># Train set<br>train_generator = ImageDataGenerator(rescale=1./255).flow_from_directory(train_dir,                                                                           <br>                        target_size = (64, 64), batch_size= 7685)</pre><pre># Validation Set<br>val_generator = ImageDataGenerator(rescale=1./255).flow_from_directory(val_dir,                                                    <br>                           target_size=(64, 64), batch_size=16)</pre><pre># Test Set<br>test_generator = ImageDataGenerator(rescale=1./255).flow_from_directory(test_dir,                                                            <br>                         target_size = (64, 64), batch_size=624)</pre><pre># Create the image-label datasets<br>train_images, train_labels = next(train_generator)<br>val_images, val_labels = next(val_generator)<br>test_images, test_labels = next(test_generator)</pre><p>The final preprocessing step is to resize the labels to an (l x 1) array, then preview an image and label to ensure that the correct labels are corresponding with the respective images.</p><pre>train_labels_final = np.reshape(train_labels[:,0], (7685,1))<br>val_labels_final = np.reshape(val_labels[:,0], (16,1))<br>test_labels_final = np.reshape(test_labels[:,0], (624,1))</pre><pre># Sanity check on image 4000<br>array_to_img(train_images[4000])<br>train_labels_final[4000, :]</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/320/1*lGGxHkhNWvV3ZhYcducqpA.png" /><figcaption>Preview of sample image 4000, an augmented image.</figcaption></figure><p>Now on to building the model. The initial model was run with the default settings, then expounded upon, after each iteration to see which parameters helped the model to achieve optimal performance. Below is a code snippet that iterates through different optimizers while fitting to the training and validation data. After the code block is run, you can visually inspect the performance of the optimizers via a line plot.</p><pre># Create a dictionary of optimizers<br>optimizers = {&quot;RMSprop&quot;: {&quot;optimizer&quot;: RMSprop(), &quot;color&quot;:&quot;blue&quot;},<br>              &quot;adam_01&quot;: {&quot;optimizer&quot;: Adam(lr=0.01),&quot;color&quot;:&quot;red&quot;},<br>              &quot;sgd&quot;: {&quot;optimizer&quot;: SGD(), &quot;color&quot;:&quot;purple&quot;},<br>              &quot;adadelta&quot;: {&quot;optimizer&quot;: Adadelta(), &quot;color&quot;:&quot;pink&quot;},<br>              &quot;adagrad&quot;: {&quot;optimizer&quot;: Adagrad(), &quot;color&quot;:&quot;yellow&quot;}}</pre><pre>for optimizer, d in optimizers.items():<br>    print(f&#39;Testing {optimizer}&#39;)<br>    <br>    # Build the CNN<br>    <br>    model = Sequential()</pre><pre>    model.add(layers.Conv2D(32, (3, 3), activation=&#39;relu&#39;,    <br>    input_shape=(64, 64, 3)))</pre><pre>    model.add(layers.MaxPooling2D((2, 2)))<br>    model.add(layers.Conv2D(32, (4, 4), activation=&#39;relu&#39;))<br>    model.add(layers.MaxPooling2D((2, 2)))</pre><pre>    model.add(layers.Conv2D(64, (3, 3), activation=&#39;relu&#39;))<br>    model.add(layers.MaxPooling2D((2,2)))</pre><pre>    # Flattening<br>    model.add(layers.Flatten())<br>    model.add(layers.Dense(64, activation = &#39;relu&#39;))<br>    model.add(layers.Dense(1, activation=&#39;softmax&#39;))<br>     model.compile(optimizer=d[&#39;optimizer&#39;],loss=&#39;binary_crossentropy&#39;,    <br>   metrics=[&#39;accuracy&#39;])</pre><pre>  results = model.fit(train_images, train_labels_final,   <br>  epochs=15, batch_size=500, validation_data=(val_images,   <br>  val_labels_final))</pre><pre>  print(&quot;Stored Results&quot;)<br>  d[&#39;loss&#39;] = history.history[&#39;loss&#39;] <br>  print(&#39;=&#39;*125) # Add a partition in between optimizer results</pre><p>Next, evaluate the final loss and accuracy of the model, then make predictions on the validation images. Access the total performance of the model using a confusion matrix in conjunction with a Classification Report. My final model resulted in having zero False Negatives and eight False Positives, which demonstrates a best case scenario because it’s better to flag a patient for additional testing, than to release a patient carrying pneumonia and their symptoms worsen or they spread viral pneumonia to a loved one.</p><pre># Evaluate the model on the metrics accuracy and loss<br>model.evaluate(train_images, train_labels_final)<br>model.evaluate(val_images, val_labels_final)</pre><pre># Make predictions on the validation set<br>preds = model.predict(val_images)<br>plot_confusion_matrix(val_labels_final, preds)</pre><pre># Print the Classification Report, which returns Recall, Precision, and F1-Score<br>classification_report =classification_report(val_labels_final, preds)</pre><p>Continue to tune your hyper parameters, and train your best performing model on your testing data, then repeat. Compare your results to a pre-trained model. Below I used <strong>VGG16.</strong></p><pre># Import Pretrained Model<br>from keras.applications.vgg16 import preprocess_input<br>from keras.applications.vgg16 import VGG16</pre><pre># Edit the default image size to (64, 64, 3)<br>pretrain_mod = VGG16(include_top= False, input_shape=(64,64,3))</pre><pre># Initialize a model<br>model = Sequential()<br>model.add(pretrain_mod)<br>model.add(layers.Flatten())<br>model.add(layers.Dense(132, activation=&#39;tanh&#39;))<br>model.add(layers.Dense(1, activation=&#39;softmax&#39;))</pre><pre># Freeze layers (do this to speed up <br>pretrain_mod.trainable = False<br>for layer in model.layers:<br>    print(layer.name, layer.trainable)<br>print(len(model.trainable_weights))</pre><pre>model.compile(loss=&#39;binary_crossentropy&#39;, optimizer=Adam(lr=0.01), metrics=[&#39;accuracy&#39;])</pre><pre>results = model.fit(train_images, train_labels_final, epochs=11, batch_size=750, validation_data=(test_images, test_labels_final))</pre><p>Don’t forget to save your best model!</p><pre>print(&#39;saving model to disk \n&#39;)<br>mod = &#39;.//Models/pretrained_mod&#39;<br>model.save(mod)</pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4e3592de208b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/computer-vision-and-pneumonia-detection-part-1-technical-4e3592de208b">Computer Vision and Pneumonia Detection Part 1: Technical</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An Analysis of “Reasonable Suspicion”]]></title>
            <link>https://medium.com/@gabriel.erica3/an-analysis-of-reasonable-suspicion-e74482438aec?source=rss-6f1460ce25f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/e74482438aec</guid>
            <category><![CDATA[social-impact]]></category>
            <category><![CDATA[police-reform]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[blacklivesmatter]]></category>
            <dc:creator><![CDATA[Erica Gabriel]]></dc:creator>
            <pubDate>Sun, 12 Jul 2020 23:09:59 GMT</pubDate>
            <atom:updated>2020-07-12T23:09:59.075Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/514/1*c5OhBp_3wxdaSCk2ZBOoGA.jpeg" /></figure><p><strong>Terry vs. Ohio (1967)</strong> the Supreme Court held that the search undertaken by the officer involved was reasonable because he <strong>“acted on more than just a hunch”, </strong>under the Fourth Amendment, which protects people against unreasonable search and seizure. Of the 43,077 people who were subjected to Terry Stops in Seattle, WA from 2015–2020, 10,505 of them were found to be carrying weapons. 9,474 subjects were frisked, and 2,173 subjects were arrested. This raises the question:</p><ul><li><em>Why were the other 33,603 stopped?</em></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bmKemOv4SBYGr9Rq1WNI8g.png" /></figure><p>The top reported 911 calls and reason documented for “Onview (“officer just witnessed an incident”) was for a “Suspicious Person”. Diving into the data we are able to see the Terry Stop Subject and reporting officer demographics below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LKTbh_cUJACa7JRwVdin-A.png" /></figure><p>African Americans are 6.1% of Seattle’s population, however they make up nearly ⅓ of those stopped. Meanwhile 49% of Seattle’s population is women, and women only make up less than ⅕ of the stops. Going back to the Call Types marked as “Suspicious Person”, the first questions that come to mind are:</p><ul><li><em>How is “suspicious” defined by police departments? By residents? Is this criteria measurable and can it be trained on?</em></li><li><em>Is the “suspicion criteria” passed on to 911 dispatchers?</em></li><li><em>Is this criteria universal? Does suspicion look the same across all races, genders, and age groups?</em></li><li><em>With considerably less women being stopped than their male counterparts, it brings up the next question, Are women generally seen as less threatening by our society? Is this a blindspot?</em></li><li><em>Are officers trained to spot when an individual is having mental health issues, and how to properly deal with that person, so that he or she can get the help that he or she needs?</em></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/366/1*tpf2twzbaBW_B9pOaxeQOg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/276/1*cEy-o9tO_kN9obJYD4-r0Q.png" /></figure><p>In the landmark Terry vs. Ohio case, Supreme Court justices ruled that the officer on duty acted on more than just a “hunch” when he stopped to search the subjects. With only 24% of the subjects discovered to be carrying a weapon. When officers are in the field either responding to 911 calls and initiating stops;</p><ul><li><em>At what point are officers operating on “more than just a hunch”? Is this instinct or chance?</em></li><li><em>How much of that instinct is based on experience in the field, or on world-views shaped by upbringing?</em></li><li><em>Are there measures put in place to prevent instincts from morphing into biases?</em></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bmKemOv4SBYGr9Rq1WNI8g.png" /></figure><p>To lessen the probability of bias fueling instincts, 911 Dispatchers and officers should be enrolled in annual implicit bias training. This is pertinent for dispatchers because as the messengers and classifiers of 911 Calls, they should be trained in asking the proper questions to identify whether actual threats are imminent, or if bias is the culprit. These fielding questions will save officers time and prevent violence against pedestrians who are not carrying weapons or committing a crime. In addition officers should be partnered with counselors who are licensed to identify individuals who are dealing with mental health issues and abuse victims. In order for community policing to be effective, the police force should at a minimum be reflective of the communities that they are serving. Specifically, minority representation within the law will help initially well intentioned laws from unfairly targeting People of Color. Investments in the areas of recruitment and soft skills training will offset the costs of lawsuits such as these:</p><ul><li><a href="https://www.mhb.com/news/seattle-settles-lawsuit-by-man-shot-in-face-by-police">Seattle Settles Lawsuit by Man Shot in Face by Police | MacDonald Hoague &amp; Bayless, Seattle Civil Rights and Immigration Attorneys</a></li><li><a href="https://budgeandheipt.com/news-press/attorneys-police-shooting-2-5-million-settlement/">https://budgeandheipt.com/news-press/attorneys-police-shooting-2-5-million-settlement/</a></li><li><a href="https://www.bellevuereporter.com/news/2-25-million-settlement-reached-in-king-county-shooting-lawsuit/">https://www.bellevuereporter.com/news/2-25-million-settlement-reached-in-king-county-shooting-lawsuit/</a></li><li><a href="https://komonews.com/news/local/city-pays-175-million-to-settle-lawsuit-over-violent-spd-arrest">https://komonews.com/news/local/city-pays-175-million-to-settle-lawsuit-over-violent-spd-arrest</a></li><li><a href="https://www.king5.com/article/news/local/protests/black-lives-matter-group-sues-over-seattle-police-tactics/281-48486b74-c897-4fba-b8e1-cebe03a3aeb8">https://www.king5.com/article/news/local/protests/black-lives-matter-group-sues-over-seattle-police-tactics/281-48486b74-c897-4fba-b8e1-cebe03a3aeb8</a></li><li><a href="https://www.thestranger.com/slog/2020/02/10/42791505/city-of-seattle-sues-county-to-block-investigations-into-fatal-police-shootings">City of Seattle Sues County to Block Investigations into Fatal Police Shootings — Slog — The Stranger</a></li><li><a href="https://apnews.com/6af77180fdc74c75906b8bf10f389db3">https://apnews.com/6af77180fdc74c75906b8bf10f389db3</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e74482438aec" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[9 Ways Companies Can Be Actionable in the Black Lives Matter Movement]]></title>
            <link>https://medium.com/@gabriel.erica3/9-ways-companies-can-be-actionable-in-the-black-lives-matter-movement-83e9e2058821?source=rss-6f1460ce25f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/83e9e2058821</guid>
            <category><![CDATA[blacklivesmatter]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[black-in-tech]]></category>
            <category><![CDATA[culture]]></category>
            <category><![CDATA[activism]]></category>
            <dc:creator><![CDATA[Erica Gabriel]]></dc:creator>
            <pubDate>Wed, 10 Jun 2020 21:30:21 GMT</pubDate>
            <atom:updated>2020-06-11T04:10:57.235Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1008/1*9iCx0JIg-G2_VBqnFn0v2A.jpeg" /></figure><p>By now we’ve all seen the “We at *insert company here* are committed to <strong>Diversity &amp; Inclusion</strong>…we stand in solidarity with the victims of police brutality…” PR templates, but what’s missing from these efforts are clear plans of action. <strong>Dismantling systemic racism involves more than just calling out police brutality; that is only one factor. Exclusive company cultures, implicit biases, hiring discrimination, are other factors</strong> where companies at all stages can use their resources to be actionable about systemic racism in the workplace. Starting with:</p><p><strong>Culture: Culture is based on WHO you hire, not kombucha on tap and ping pong tables.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/559/1*F6YQ9GxpsekvSjXgRHKjlQ.jpeg" /></figure><ol><li>It’s <strong>“Black”</strong>, not <strong>“black”</strong>, and <strong>“People of Color”</strong> is not a synonym for Black People. It is an erasure. Update your publications accordingly.</li><li>Stop having White People conduct <strong>Social Justice</strong>, <strong>Diversity &amp; Inclusion</strong>, and <strong>Racial Sensitivity Training</strong>. I don’t think I need to explain why this shouldn’t be a thing.</li><li>The <strong>HR Department</strong> needs to evolve. Their past role of preventing companies from being sued perpetuates <strong>racism and sexism</strong> by forcing out employees (majority Black women) with valid grievances and rewarding the “<a href="https://news.sky.com/video/amy-cooper-woman-calls-cops-on-african-american-man-in-new-york-11995375">Amy Cooper’s</a>” and “<a href="https://www.kansascity.com/news/nation-world/national/article243030121.html">Tom Austin’s</a>” in your office with career longevity. More emphasis should be placed on <strong>repairing</strong> <strong>your company culture</strong>, instead of covering it up, then you wont have any discrimination lawsuits.</li></ol><p><strong>Leadership</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5UUnn0Qe1oeQrTcLubdz_g.jpeg" /></figure><ol><li>Hire and promote Black People to your <strong>Executive Team</strong> and <strong>Board of Directors</strong>. You can’t truly be committed to Diversity &amp; Inclusion if your executive team looks like a pack of Mark Zuckerburgs and Ann Coulters.</li><li><strong>Black people need to be taken seriously as SMEs in areas beyond D &amp; I.</strong> It’s everyone’s job to uphold diversity and inclusion, and should your work environment not reflect that, then that problem falls on everyone’s shoulders, not the one Black guy who has a PhD in neuroscience that you hired to plan MLK Day community service events and movie nights during Black History Month. Put Black people in positions where they can directly <strong>influence the technology</strong>, and <strong>impact business &amp; legal decisions</strong>. Allow them to use their skills and credentials, this will only make your company and the products produced stronger.</li><li>Black employees are often over mentored and under promoted. All employees need to be provided with a <strong>clear roadmap to promotion</strong>, and grant them more <strong>opportunities for career visibility</strong>, so that upper management will be able to vouch for them when discussion of promotions and raises occur.</li></ol><p><strong>Recruitment and Hiring</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WaGab67hZ0atYdD09jn1uw.jpeg" /></figure><ol><li>When recruiting at colleges, if you are recruiting at Predominately White Institutions (PWIs), reach out to the <strong>Black Student Unions</strong> there to set up information sessions, resume building workshops, and find out how your company can participate in their career fairs if applicable. Then do the same for the <strong>Historically Black Colleges and Universities </strong>(HBCUs) in the surrounding areas. This goes without saying, but make sure the recruiters are warm and inviting. If your recruiters are stand-offish, or racist <strong>*cough cough*,</strong> it will only give potential candidates a glimpse of what they can expect if you actually worked for your company, therefore they will not visit your booth.</li><li>Hire Black recruiters.</li><li>Make sure that the <strong>non-White person</strong> who is conducting any inclusion trainings, includes <strong>implicit bias</strong> in their training module, so that employees can confront their biases head on and check themselves when they say that a Black candidate is <strong>“not a culture fit”</strong>, but they “can’t quite put their finger on it”.</li></ol><p>Now that you <strong>know</strong> better, it’s time to <strong>do</strong> better. Let your social media activism be purposeful, not performative.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/271/1*tzdEbPUZkGK9sAm3U7LtFA.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=83e9e2058821" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Life Gives You Cauliflower, Make Cauliflower Wings]]></title>
            <link>https://medium.com/when-life-gives-you-cauliflower-make-cauliflower/when-life-gives-you-cauliflower-make-cauliflower-wings-a3316f932b61?source=rss-6f1460ce25f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/a3316f932b61</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[women-in-tech]]></category>
            <dc:creator><![CDATA[Erica Gabriel]]></dc:creator>
            <pubDate>Fri, 08 May 2020 04:38:18 GMT</pubDate>
            <atom:updated>2020-05-08T04:38:18.765Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve always known what I wanted. At 13 I decided that I wanted to pursue Mechanical Engineering, and at 16, I decided that I had to work for myself. I had gotten my first job, and was working 10–12 hours a shift. Doing grueling, menial tasks all day for a company that I didn’t care about, who’s mission I did not believe in. It motivated me to spend the remaining years of high school looking beyond what I would major in, and looking into companies where I can begin my career. Only I couldn’t find a company who was focusing on the social issues that I wanted to solve. I’d try to concoct ways of how I could persuade them to pivot in my direction, but the amount of red tape involved in raising a new idea or project, was just pure ridiculousness. I just couldn’t imagine spending my entire working career burning myself out to make someone else’s dream come true. I needed to be my own boss.</p><p>Fast forward to undergrad at Georgia Tech, I discovered the term “entrepreneurship’ and dove head first into Atlanta’s startup ecosystem. That experience ultimately solidified that entrepreneurship was what I was seeking and wanted to pursue to make my vision come into fruition. My last few years of undergrad, I was placing in hack-a-thons, social entrepreneurship case competitions, and I even interned at a startup! When the time came to begin my full-time job search, naturally everyone’s asking “What are you gonna do after graduation?” And I would answer excitedly, “I’m gonna go into entrepreneurship! I want to-”, as if a gut check reaction, people would immediately hit me with the “But what if…”, “But what about…?”, “You need a job that’s going to get you a guaranteed salary and benefits.”, or my personal favorite “Here’s all the bad things that are going to happen with startups that are 100% true because I read them in a facebook article.” I was getting it from all directions, and I must admit, points were made. So I reacted and changed my career trajectory to the “safer option” and focused on traditional careers at Fortune 500 Companies, and said that I would revisit the startup thing in the future after I had built up a nest egg.</p><p>At my first employer, I took on a Project Management Role, and was overworked and over utilized. I was burned out to the point where I didn’t have the time or the energy to work on my passions outside of work, so I left. Immediately after I went to work for an early stage startup and was in an operations, project management type of role, but I was stifled, and vastly underutilized. I managed to gain some key takeaways from these companies such as, how to be “scrappy”; products don’t need to be perfect to launch, just get it out there and iterate as you go.I learned how to delegate tasks, how to create processes, communicate with non technical partners, and the benefits of long form communication. By far the largest lessons I took away from these experiences was learning the type of boss I wanted and did not want to be, and how to create culture, and no it is not kombucha and beer on tap, guys.<strong> </strong>However, it felt like my technical skills were slipping away from me. There were so many new technologies that had emerged since I graduated like Machine Learning, Data Science, Deep Learning, etc. that I wanted to get into, but there are some skills that you can’t “job shadow” your way into. After being unhappy and complacent for months, the job ended. And shockingly, I was relieved. The universe had given me an out, and I took it. I realized that I kept trying to squeeze myself into these “safe” spaces where my dreams and interests did not fit. Those family and friends that I had spoken to pre-graduation had the best of intentions, but none of them had ANY entrepreneurship experience. I mean, they hadn’t even stepped inside of the building of a startup. So how were they so firm in their convictions? I reanalyzed each of those conversations and realized that the majority of those folks I had spoken to had a pattern of being risk averse; while I love them, why was I telling them about my big hairy goals? WRONG AUDIENCE. They were just projecting their fears onto me; it’s what we as people do, it’s what everyone does, but this time I wasn’t going to let it affect me.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MEpzHKCpEbef_XjZ126zwg.jpeg" /><figcaption>First Day of School!</figcaption></figure><p>I promised myself that I would use this opportunity to realign myself with my earlier career goals; using technology to solve social issues. Seeing that the world was moving in the direction of machine learning and automation, and taking a deep look at the founders of companies who were pioneering this technology, I couldn’t help but notice that these founders resembled your typical tech bros. My spidey senses told me that people of color were sure to be excluded from quality of life improving technical solutions. Solutions that could provide access and affordability to fresh food, education, professional careers, and mental health care. This led me to begin researching how I could reduce bias in machine learning algorithms and what softwares I needed to learn to make it happen. I started looking at alternative learning programs because I was NOT checking for another 5 years in school and stumbled upon Flatiron School’s Data Science Program. Fast forward to now, the experience has been amazing so far, it’s like drinking from a fire hose and the program moves faster than anything I’ve ever done before, but I’m loving every second of it. I don’t regret making this investment in myself at all. Now why do I share all of this? Is this one of those “Quit your job, start a business, get evicted, sleep in your car, and have sleep for dinner!” blog posts? Or one of those Everett College style commercials in written form; “What are you doing with your life?! Nothing!, sitting on the couch, being a trash can, empty your life savings and do something with yourself today!” Nope, quite the opposite.</p><p>By now with all of the COVID-19 related layoffs, I think the world is learning the same lesson: There are no “safe” or stable jobs, companies, or careers. You can get laid off from something you hate; you can fail at something you hate. If you’ve got the bandwidth, you owe it to yourself to try at something you love. Choose your struggle. While you’re sitting at home submitting 50 applications a day to companies you won’t remember, or sitting through that 30 minute phone screen only to hear “We’re actually looking for someone who has “<em>these skills</em>” *insert skills you don’t have here*.” Take a moment to get curious. Figure out what it is that you actually want to do, and what’s the fastest and cheapest way to get those skills. For me it was a coding bootcamp, but for you it could be entirely different. Don’t sleep on those $12 courses on Udemy, or the free content that Youtube provides; whatever you do just don’t fall back into the cycle of complacency and settling. You’ve got gifts the rest of the world needs to see. I’m willing to bet that you’ve also got a few risk averse people in your circle; you don’t have to tell <em>everyone </em>your dreams and plans. The amount of unwarranted advice is enough to make your head pop off of your shoulders and roll over into a ditch. Just pull up a Google Doc, drop your plan of attack in there, and keep it pushing. Never give up your search to find something that gets you out of bed in the morning, and put yourself in the position(s) to get it. If no one else is rooting for you, I am.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a3316f932b61" width="1" height="1" alt=""><hr><p><a href="https://medium.com/when-life-gives-you-cauliflower-make-cauliflower/when-life-gives-you-cauliflower-make-cauliflower-wings-a3316f932b61">When Life Gives You Cauliflower, Make Cauliflower Wings</a> was originally published in <a href="https://medium.com/when-life-gives-you-cauliflower-make-cauliflower">When Life Gives You Cauliflower, Make Cauliflower Wings</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>