zeryx

package com.algorithmia

import com.algorithmia.handler.AbstractAlgorithm
import ml.combust.bundle.BundleFile
import ml.combust.bundle.dsl.Bundle
import ml.combust.mleap.core.types._
import ml.combust.mleap.runtime.MleapSupport._
import ml.combust.mleap.runtime.frame.{DefaultLeapFrame, Row, Transformer}

import scala.collection.mutable

zeryx

What are we going to do today?

• We'll understand what this gitlab -> algorithmia integration does
• Gitlab -> Algorithmia Procedure:
  • Create a new Algorithm on Algorithmia
  • Create a new project in Gitlab
  • Add our secret variables to the GitLab project from Algorithmia
  • Clone both git repositories to our local system
• Copy over template code from the Algorithmia repo to the Gitlab repo

zeryx

[{ "ClaimId ":5355879, "CciliId ":9807378, "PetPolicyId ":1019191, "OwnerId ":847019, "Ded ":250.0000, "RiderInEffect ":None, "DaysInEffect ":None, "dateclosed ":None, "Inserted ": "2020-04-29T06:23:44.277 ", "DateOfLoss ": "2020-04-29T04:00:00 ", "DaysEnrollLoss ":1196, "ItemDescription ": "DALLAS:  EXAMINATION- RECHECK ", "TrialFlag ":1, "PolicyPetName ": "DALLAS ", "InvoicePetName ": "DALLAS ", "DiagnosisOne ": "DIAGNOSTICS ", "DiagnosisTwo ": " ", "PawprintStatus ": "NoPrex ", "SalesTaxFlag ":0, "Subtotal ":113.7500, "ActualExtendedPrice ":36.0000, "DiscountFlag ":0, "PaymentFlag ":0, "LineItemDupFlag ":0, "DuplicateClaimId ":None, "TaxCovered ":0, "Eligible ":0.0, "PastCondList ": "ABRASION,GASTROENTERITIS,RETAINED DECIDUOUS TOOTH ", "TransactionDate ": "2020-04-29T04:00:00 ", "EnrollmentDate ": "2017-01-31T20:12:59.857 ", "TrialDate ": "2017-01-19T15:25:15.203 ", "FirstSymptomDate ": "2020-04-15T04:00:00 ", "RelatedClaim ":5314280, "AgeDays ":1196, "PolicyStartDate ": "2017-01-19T00:00:00 "},{ "ClaimId 

zeryx

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torchvision import datasets, models, transforms
import time

# Binary Classification example using a pretrained imagenet model

zeryx

import torch as th
from torchvision import models, transforms
from PIL import Image
import numpy as np
resnet18 = models.resnet18(pretrained=True)
# This stage below removes the last module (aka the softmax layer) to provide raw data for fine-tuning/transfer learning
modules=list(resnet18.children())[:-1]
model = th.nn.Sequential(*modules)

th.save(model, "/tmp/resnet18-headless.th")

zeryx

algorithmia>=1.0.0,<2.0
six
tensorflow-gpu==1.2.0

zeryx

Steps required to deploy a new dependency package

Langpacks

1. Create new langpacks branch, from latest master
   • eg: git checkout -b ALGO-xxx-package-name
2. create new directory in libraries
   • eg: package-name-version
3. create new Dockerfile in new package directory
4. copy existing dependency template to new directory

• eg: cp -LrR templates/package-name-version-1 templates/package-name-version

zeryx

Already done

• ResNet 18 - Image Classification - ImageNet (not available for tensorflow)
• ResNet 50 - Image Classification - ImageNet

Suggestions

• MobileNet v2 - Image Classification
• Why

zeryx

import tensorflow as tf
import Algorithmia
import os
import re
import numpy as np


TEMP_COLLECTION = 'data://.session/'
MODEL_FILE = "data://zeryx/InceptionNetDemo/classify_image_graph_def.pb"
CONVERSION_FILE = "data://zeryx/InceptionNetDemo/imagenet_synset_to_human_label_map.txt"

zeryx

import Algorithmia
from .auxillary import load_model, id_to_string
import tensorflow as tf
import asyncio
import os, re
import numpy as np


# Global variables here
client = Algorithmia.client()