import {z} from 'zod'

export type GetApiApiProjectIdPretrainedModelArgs = {projectId: number, impulseId?: number | undefined};

export type GetApiApiProjectIdPretrainedModelOptions = {baseUrl?: string | undefined, ApiKeyAuthentication?: string | undefined, fetch?: typeof fetch | undefined};

export const kerasModelTypeEnum = z.enum(["int8", "float32", "akida", "requiresRetrain"]);

export const kerasModelVariantEnum = z.enum(["int8", "float32", "akida"]);

export const profileModelInfoMemoryDetails = z.object({ram: z.number().int(), rom: z.number().int(), arenaSize: z.number().int()});

export const kerasCustomMetric = z.object({name: z.string(), value: z.string()});

export const profileModelInfo = z.object({variant: kerasModelVariantEnum, device: z.string(), tfliteFileSizeBytes: z.number().int(), isSupportedOnMcu: z.boolean(), memory: z.object({tflite: profileModelInfoMemoryDetails.optional(), eon: profileModelInfoMemoryDetails.optional(), eonRamOptimized: profileModelInfoMemoryDetails.optional()}).optional(), timePerInferenceMs: z.number().int().optional(), mcuSupportError: z.string().optional(), customMetrics: z.array(kerasCustomMetric), hasPerformance: z.boolean(), profilingError: z.string().optional()});

export const profileModelTableMcu = z.object({description: z.string(), timePerInferenceMs: z.number().int().optional(), memory: z.object({tflite: z.object({ram: z.number().int(), rom: z.number().int()}).optional(), eon: z.object({ram: z.number().int(), rom: z.number().int()}).optional(), eonRamOptimized: z.object({ram: z.number().int(), rom: z.number().int()}).optional()}).optional(), supported: z.boolean(), mcuSupportError: z.string().optional()});

export const profileModelTableMpu = z.object({description: z.string(), timePerInferenceMs: z.number().int().optional(), rom: z.number().optional(), supported: z.boolean()});

export const profileModelTable = z.object({variant: z.enum(["int8", "float32"]), lowEndMcu: profileModelTableMcu, highEndMcu: profileModelTableMcu, highEndMcuPlusAccelerator: profileModelTableMcu, mpu: profileModelTableMpu, gpuOrMpuAccelerator: profileModelTableMpu});

export const pretrainedModelTensor = z.object({dataType: z.enum(["int8", "uint8", "float32"]), name: z.string(), shape: z.array(z.number().int()), quantizationScale: z.number().optional(), quantizationZeroPoint: z.number().optional()});

export const deployPretrainedModelInputTimeSeries = z.object({inputType: z.literal("time-series"), frequencyHz: z.number(), windowLengthMs: z.number().int()});

export const deployPretrainedModelInputAudio = z.object({inputType: z.literal("audio"), frequencyHz: z.number()});

export const imageInputScaling = z.enum(["0..1", "-1..1", "-128..127", "0..255", "torch", "bgr-subtract-imagenet-mean"]);

export const imageInputResizeMode = z.enum(["squash", "fit-short", "fit-long", "crop"]);

export const deployPretrainedModelInputImage = z.object({inputType: z.literal("image"), inputScaling: imageInputScaling.optional(), resizeMode: imageInputResizeMode.optional()});

export const deployPretrainedModelInputOther = z.object({inputType: z.literal("other")});

export const deployPretrainedModelModelClassification = z.object({modelType: z.literal("classification"), labels: z.array(z.string()), addSoftmax: z.boolean().optional()});

export const deployPretrainedModelModelRegression = z.object({modelType: z.literal("regression")});

export const objectDetectionLastLayer = z.enum(["mobilenet-ssd", "fomo", "yolov2-akida", "yolov5", "yolov5v5-drpai", "yolox", "yolov7", "yolo-pro", "tao-retinanet", "tao-ssd", "tao-yolov3", "tao-yolov4", "yolov11", "yolov11-abs", "paddleocr-detector", "qc-face-det-lite", "qc-yolox"]);

export const deployPretrainedModelModelObjectDetection = z.object({modelType: z.literal("object-detection"), labels: z.array(z.string()), lastLayer: objectDetectionLastLayer, minimumConfidence: z.number(), thresholdValues: z.record(z.string(), z.number()).optional()});

export const deployPretrainedModelModelFreeform = z.object({modelType: z.literal("freeform")});

export const deployPretrainedModelModelAnomaly = z.object({modelType: z.literal("anomaly")});

export const thresholdValue = z.union([z.number(), z.string()]);

export const deployPretrainedModelModelVisualAnomaly = z.object({modelType: z.literal("visual-anomaly"), thresholdValues: z.record(z.string(), thresholdValue).optional()});

export const getPretrainedModelResponse = z.object({success: z.boolean(), error: z.string().optional(), specificDeviceSelected: z.boolean(), availableModelTypes: z.array(kerasModelTypeEnum), model: z.object({fileName: z.string(), profileInfo: z.object({float32: profileModelInfo.optional(), int8: profileModelInfo.optional(), table: profileModelTable}).optional(), inputs: z.array(pretrainedModelTensor), outputs: z.array(pretrainedModelTensor), profileJobId: z.number().int().optional(), profileJobFailed: z.boolean().optional(), supportsTFLite: z.boolean().optional()}).optional(), modelInfo: z.object({input: z.discriminatedUnion("inputType", [deployPretrainedModelInputTimeSeries, deployPretrainedModelInputAudio, deployPretrainedModelInputImage, deployPretrainedModelInputOther]), model: z.discriminatedUnion("modelType", [deployPretrainedModelModelClassification, deployPretrainedModelModelRegression, deployPretrainedModelModelObjectDetection, deployPretrainedModelModelFreeform, deployPretrainedModelModelAnomaly, deployPretrainedModelModelVisualAnomaly])}).optional()});

export const validateResponse = async <Output>(
  schema: {
    readonly '~standard': {
      readonly validate: (value: unknown) => ({ readonly value: Output; readonly issues?: undefined } | { readonly issues: ReadonlyArray<{ readonly message: string; readonly path?: ReadonlyArray<PropertyKey | { readonly key: PropertyKey }> | undefined }> }) | Promise<{ readonly value: Output; readonly issues?: undefined } | { readonly issues: ReadonlyArray<{ readonly message: string; readonly path?: ReadonlyArray<PropertyKey | { readonly key: PropertyKey }> | undefined }> }>
    }
  },
  value: unknown
): Promise<Output> => {
  const result = await schema['~standard'].validate(value)

  if (result.issues !== undefined) {
    const detail = result.issues
      .map(({ message, path }) =>
        path?.length
          ? `${path.map((segment) => String(typeof segment === 'object' ? segment.key : segment)).join('.')}: ${message}`
          : message
      )
      .join('; ')

    throw new Error(`Response does not match the schema: ${detail}`, { cause: { issues: result.issues } })
  }

  return result.value
};

/**
 * Get pretrained model
 * 
 * GET /api/{projectId}/pretrained-model
 * 
 * Also accepts auth: JWTAuthentication, JWTHttpHeaderAuthentication, OAuth2
 */
export const getApiApiProjectIdPretrainedModel = async (args: GetApiApiProjectIdPretrainedModelArgs, options?: GetApiApiProjectIdPretrainedModelOptions) => {
const {baseUrl = "https://studio.edgeimpulse.com/v1", fetch: fetchFn = fetch, ApiKeyAuthentication} = options ?? {}

const query = new URLSearchParams()
const headers = new Headers()
headers.set('accept', "application/json")

if (args.impulseId != null) query.append("impulseId", String(args.impulseId))
if (ApiKeyAuthentication !== undefined) headers.set("x-api-key", ApiKeyAuthentication)

const url = `${baseUrl.replace(/\/+$/, '')}/api/${encodeURIComponent(String(args.projectId))}/pretrained-model${query.size > 0 ? `?${query}` : ''}`

const res = await fetchFn(url, {method: "GET", headers})

if (!res.ok) throw new Error(`Request failed with status ${res.status}`, {cause: {status: res.status, body: await res.text()}})

return validateResponse(getPretrainedModelResponse, await res.json())
};