Related Posts
Does Walmart uses dell boomi
Guys, a query from me - got an offer from a reputed renewable industry company so wanted to know if I'll be going downhill or uphill in my career. Note: Position is in Europe, naturally way better money than India, and its an ex client wanting me to join them. He is keen to get me in.
https://joinfishbowl.com/post_1ysj22yk7e
Mentor
Conversation Starter
Hello,
Received an verbal offer from Amazon for Data Engineer position (Luxembourg).
I have no idea regarding pay bands for Amazon Luxembourg.
Wanted to check if there is any room for improvement with respect to proposed offer.
Base : 85k €
Bonus : 1st year: 14k and 2nd year : 12k €
RSU : 24 Units
YoE : 4
TC : 42 LPA
Amazon
Anyone got salary today??
Conversation Starter
Rising Star
Additional Posts in Data Science and Machine Learning
Hey Folks,
Help me to unlock DM!
Thanks.
Hi, I recently came across Scaler's MS program, which offers masters degree in collaboration with Woolf University (European University). And, I am looking move to Canada in near future. Will the master degree offered through this course will be accepted by the education evaluation institute WES? If yes, is it considered as equivalent to Canadian Masters degree? Scaler Academy Immigration and Refugee Board of Canada Immigration and Refugee Board of Canada
New to Fishbowl?
unlock all discussions on Fishbowl.
Community Builder
My approach is to start with simple Architecture and check if there any better model or architecture available for problem statement we are solving and Experiment by adding each layer or additional layers and Logging these Model Evaluation Metrics and whichever gives u the best Results that's the best Model.
Accuracy is not the best Metric always,it depends on what type of Problem statement you are solving.
Let's say we are doing Health Disease classification and In which I want the model to predict all the cases of the Disease ie, Recall should be Higher because even though model Classified Normal instance as Disease instance ,those patients can check again or do precautions tests to confirm again but if u left any classify any diseased patients as Normal that's gonna create chaos.
Normally researchers prefer F1 score as metric and also do cross Validation to check Model Accurate Results.
And Plot different Metrics plots like Roc ,Prc curve those Will give u some extra idea about which model u should choose.
I think u should record whatever custom layers u are adding and the Parameters of it and results of it and combinations of it for logging ,this way u can easily find out which method or combination is giving u best results.
A lot depends on your input data type, structure and size. If data is all binary then a tree based model might be a bad choice and if it is an image then a CNN/Transformer based model should be chosen. Also, depending on the number of data points you have for training one might decide if a particular architecture is chosen, will it even converge to a local minima.
To start with, these pointers help. Once you have finalized on the model types, I guess then its all about experimenting with readily available models. If you see an opportunity to further improve the model metrics then you can go ahead with a custom model architecture.
Last but not the least, the gpu compute available at your hand must be considered before you select a specific architecture. Experimentation time drastically changes based on your architecture selection.
Hope this helps!